[llvm] [AMDGPU] Extend type support for update_dpp intrinsic (PR #114597)

Stanislav Mekhanoshin via llvm-commits llvm-commits at lists.llvm.org
Mon Nov 4 22:45:54 PST 2024


https://github.com/rampitec updated https://github.com/llvm/llvm-project/pull/114597

>From 63fbcaab4055c87a759d373013230b538089bcbe Mon Sep 17 00:00:00 2001
From: Stanislav Mekhanoshin <Stanislav.Mekhanoshin at amd.com>
Date: Thu, 31 Oct 2024 12:32:12 -0700
Subject: [PATCH 1/4] [AMDGPU] Extend type support for update_dpp intrinsic

We can split 64-bit DPP as a post-RA pseudo if control values are
supported, but cannot handle other types.
---
 .../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp |   49 +-
 llvm/lib/Target/AMDGPU/SIISelLowering.cpp     |   50 +-
 .../GlobalISel/llvm.amdgcn.update.dpp.ll      |   40 +-
 .../atomic_optimizations_global_pointer.ll    | 1756 +++---
 .../atomic_optimizations_local_pointer.ll     | 5303 ++++++++---------
 llvm/test/CodeGen/AMDGPU/dpp64_combine.ll     |   16 +-
 .../AMDGPU/global_atomics_scan_fadd.ll        | 1457 +++--
 .../AMDGPU/global_atomics_scan_fmax.ll        |  926 ++-
 .../AMDGPU/global_atomics_scan_fmin.ll        |  926 ++-
 .../AMDGPU/global_atomics_scan_fsub.ll        | 1457 +++--
 .../AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll   |  258 +
 .../CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll  |   30 +-
 12 files changed, 6163 insertions(+), 6105 deletions(-)
 create mode 100644 llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index ffff34417232bb..23dbf50a0480b8 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -5495,6 +5495,13 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
     }
     case Intrinsic::amdgcn_mov_dpp8:
       return LaneOp.addImm(MI.getOperand(3).getImm()).getReg(0);
+    case Intrinsic::amdgcn_update_dpp:
+      return LaneOp.addUse(Src1)
+          .addImm(MI.getOperand(4).getImm())
+          .addImm(MI.getOperand(5).getImm())
+          .addImm(MI.getOperand(6).getImm())
+          .addImm(MI.getOperand(7).getImm())
+          .getReg(0);
     default:
       llvm_unreachable("unhandled lane op");
     }
@@ -5504,7 +5511,7 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
   Register Src0 = MI.getOperand(2).getReg();
   Register Src1, Src2;
   if (IID == Intrinsic::amdgcn_readlane || IID == Intrinsic::amdgcn_writelane ||
-      IsSetInactive || IsPermLane16) {
+      IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16) {
     Src1 = MI.getOperand(3).getReg();
     if (IID == Intrinsic::amdgcn_writelane || IsPermLane16) {
       Src2 = MI.getOperand(4).getReg();
@@ -5514,7 +5521,14 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
   LLT Ty = MRI.getType(DstReg);
   unsigned Size = Ty.getSizeInBits();
 
-  if (Size == 32) {
+  unsigned SplitSize =
+      (IID == Intrinsic::amdgcn_update_dpp && (Size % 64 == 0) &&
+       ST.hasDPALU_DPP() &&
+       AMDGPU::isLegalDPALU_DPPControl(MI.getOperand(4).getImm()))
+          ? 64
+          : 32;
+
+  if (Size == SplitSize) {
     // Already legal
     return true;
   }
@@ -5522,7 +5536,7 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
   if (Size < 32) {
     Src0 = B.buildAnyExt(S32, Src0).getReg(0);
 
-    if (IsSetInactive || IsPermLane16)
+    if (IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16)
       Src1 = B.buildAnyExt(LLT::scalar(32), Src1).getReg(0);
 
     if (IID == Intrinsic::amdgcn_writelane)
@@ -5534,31 +5548,27 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
     return true;
   }
 
-  if (Size % 32 != 0)
+  if (Size % SplitSize != 0)
     return false;
 
-  LLT PartialResTy = S32;
+  LLT PartialResTy = LLT::scalar(SplitSize);
   if (Ty.isVector()) {
     LLT EltTy = Ty.getElementType();
-    switch (EltTy.getSizeInBits()) {
-    case 16:
-      PartialResTy = Ty.changeElementCount(ElementCount::getFixed(2));
-      break;
-    case 32:
+    unsigned EltSize = EltTy.getSizeInBits();
+    if (EltSize == SplitSize)
       PartialResTy = EltTy;
-      break;
-    default:
-      // Handle all other cases via S32 pieces;
-      break;
-    }
+    else if (EltSize == 16 || EltSize == 32)
+      PartialResTy =
+          Ty.changeElementCount(ElementCount::getFixed(SplitSize / EltSize));
+    // Handle all other cases via S32/S64 pieces;
   }
 
-  SmallVector<Register, 2> PartialRes;
-  unsigned NumParts = Size / 32;
+  SmallVector<Register, 4> PartialRes;
+  unsigned NumParts = Size / SplitSize;
   MachineInstrBuilder Src0Parts = B.buildUnmerge(PartialResTy, Src0);
   MachineInstrBuilder Src1Parts, Src2Parts;
 
-  if (IsSetInactive || IsPermLane16)
+  if (IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16)
     Src1Parts = B.buildUnmerge(PartialResTy, Src1);
 
   if (IID == Intrinsic::amdgcn_writelane)
@@ -5567,7 +5577,7 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
   for (unsigned i = 0; i < NumParts; ++i) {
     Src0 = Src0Parts.getReg(i);
 
-    if (IsSetInactive || IsPermLane16)
+    if (IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16)
       Src1 = Src1Parts.getReg(i);
 
     if (IID == Intrinsic::amdgcn_writelane)
@@ -7555,6 +7565,7 @@ bool AMDGPULegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
   case Intrinsic::amdgcn_set_inactive:
   case Intrinsic::amdgcn_set_inactive_chain_arg:
   case Intrinsic::amdgcn_mov_dpp8:
+  case Intrinsic::amdgcn_update_dpp:
     return legalizeLaneOp(Helper, MI, IntrID);
   case Intrinsic::amdgcn_s_buffer_prefetch_data:
     return legalizeSBufferPrefetch(Helper, MI);
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 68ea6f622feca5..7aac584750d20f 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -6162,6 +6162,14 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
                        IID == Intrinsic::amdgcn_set_inactive_chain_arg;
   SDLoc SL(N);
   MVT IntVT = MVT::getIntegerVT(ValSize);
+  const GCNSubtarget &ST =
+      DAG.getMachineFunction().getSubtarget<GCNSubtarget>();
+  unsigned SplitSize =
+      (IID == Intrinsic::amdgcn_update_dpp && (ValSize % 64 == 0) &&
+       ST.hasDPALU_DPP() &&
+       AMDGPU::isLegalDPALU_DPPControl(N->getConstantOperandVal(3)))
+          ? 64
+          : 32;
 
   auto createLaneOp = [&DAG, &SL, N, IID](SDValue Src0, SDValue Src1,
                                           SDValue Src2, MVT ValT) -> SDValue {
@@ -6169,6 +6177,7 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
     switch (IID) {
     case Intrinsic::amdgcn_permlane16:
     case Intrinsic::amdgcn_permlanex16:
+    case Intrinsic::amdgcn_update_dpp:
       Operands.push_back(N->getOperand(6));
       Operands.push_back(N->getOperand(5));
       Operands.push_back(N->getOperand(4));
@@ -6206,13 +6215,15 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
   SDValue Src0 = N->getOperand(1);
   SDValue Src1, Src2;
   if (IID == Intrinsic::amdgcn_readlane || IID == Intrinsic::amdgcn_writelane ||
-      IID == Intrinsic::amdgcn_mov_dpp8 || IsSetInactive || IsPermLane16) {
+      IID == Intrinsic::amdgcn_mov_dpp8 ||
+      IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16) {
     Src1 = N->getOperand(2);
-    if (IID == Intrinsic::amdgcn_writelane || IsPermLane16)
+    if (IID == Intrinsic::amdgcn_writelane ||
+        IID == Intrinsic::amdgcn_update_dpp || IsPermLane16)
       Src2 = N->getOperand(3);
   }
 
-  if (ValSize == 32) {
+  if (ValSize == SplitSize) {
     // Already legal
     return SDValue();
   }
@@ -6222,7 +6233,7 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
     Src0 = DAG.getAnyExtOrTrunc(IsFloat ? DAG.getBitcast(IntVT, Src0) : Src0,
                                 SL, MVT::i32);
 
-    if (IsSetInactive || IsPermLane16) {
+    if (IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16) {
       Src1 = DAG.getAnyExtOrTrunc(IsFloat ? DAG.getBitcast(IntVT, Src1) : Src1,
                                   SL, MVT::i32);
     }
@@ -6237,7 +6248,7 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
     return IsFloat ? DAG.getBitcast(VT, Trunc) : Trunc;
   }
 
-  if (ValSize % 32 != 0)
+  if (ValSize % SplitSize != 0)
     return SDValue();
 
   auto unrollLaneOp = [&DAG, &SL](SDNode *N) -> SDValue {
@@ -6284,21 +6295,26 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
     switch (MVT::SimpleValueType EltTy =
                 VT.getVectorElementType().getSimpleVT().SimpleTy) {
     case MVT::i32:
-    case MVT::f32: {
-      SDValue LaneOp = createLaneOp(Src0, Src1, Src2, VT.getSimpleVT());
-      return unrollLaneOp(LaneOp.getNode());
-    }
+    case MVT::f32:
+      if (SplitSize == 32) {
+        SDValue LaneOp = createLaneOp(Src0, Src1, Src2, VT.getSimpleVT());
+        return unrollLaneOp(LaneOp.getNode());
+      }
+      [[fallthrough]];
     case MVT::i16:
     case MVT::f16:
     case MVT::bf16: {
-      MVT SubVecVT = MVT::getVectorVT(EltTy, 2);
+      unsigned SubVecNumElt =
+          SplitSize / VT.getVectorElementType().getSizeInBits();
+      MVT SubVecVT = MVT::getVectorVT(EltTy, SubVecNumElt);
       SmallVector<SDValue, 4> Pieces;
       SDValue Src0SubVec, Src1SubVec, Src2SubVec;
-      for (unsigned i = 0, EltIdx = 0; i < ValSize / 32; i++) {
+      for (unsigned i = 0, EltIdx = 0; i < ValSize / SplitSize; i++) {
         Src0SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, SL, SubVecVT, Src0,
                                  DAG.getConstant(EltIdx, SL, MVT::i32));
 
-        if (IsSetInactive || IsPermLane16)
+        if (IID == Intrinsic::amdgcn_update_dpp || IsSetInactive ||
+            IsPermLane16)
           Src1SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, SL, SubVecVT, Src1,
                                    DAG.getConstant(EltIdx, SL, MVT::i32));
 
@@ -6307,10 +6323,10 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
                                    DAG.getConstant(EltIdx, SL, MVT::i32));
 
         Pieces.push_back(
-            IsSetInactive || IsPermLane16
+            IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16
                 ? createLaneOp(Src0SubVec, Src1SubVec, Src2, SubVecVT)
                 : createLaneOp(Src0SubVec, Src1, Src2SubVec, SubVecVT));
-        EltIdx += 2;
+        EltIdx += SubVecNumElt;
       }
       return DAG.getNode(ISD::CONCAT_VECTORS, SL, VT, Pieces);
     }
@@ -6320,10 +6336,11 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
     }
   }
 
-  MVT VecVT = MVT::getVectorVT(MVT::i32, ValSize / 32);
+  MVT VecVT =
+      MVT::getVectorVT(MVT::getIntegerVT(SplitSize), ValSize / SplitSize);
   Src0 = DAG.getBitcast(VecVT, Src0);
 
-  if (IsSetInactive || IsPermLane16)
+  if (IID == Intrinsic::amdgcn_update_dpp || IsSetInactive || IsPermLane16)
     Src1 = DAG.getBitcast(VecVT, Src1);
 
   if (IID == Intrinsic::amdgcn_writelane)
@@ -8833,6 +8850,7 @@ SDValue SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::amdgcn_set_inactive:
   case Intrinsic::amdgcn_set_inactive_chain_arg:
   case Intrinsic::amdgcn_mov_dpp8:
+  case Intrinsic::amdgcn_update_dpp:
     return lowerLaneOp(*this, Op.getNode(), DAG);
   default:
     if (const AMDGPU::ImageDimIntrinsicInfo *ImageDimIntr =
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.update.dpp.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.update.dpp.ll
index 4b0f2ef77a9834..336a7767ee0622 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.update.dpp.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.update.dpp.ll
@@ -52,11 +52,11 @@ define amdgpu_kernel void @update_dppi64_test(ptr addrspace(1) %arg, i64 %in1, i
 ; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GFX8-NEXT:    flat_load_dwordx2 v[2:3], v[0:1]
-; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s2
+; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    v_mov_b32_dpp v4, v2 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    flat_store_dwordx2 v[0:1], v[4:5]
 ; GFX8-NEXT:    s_endpgm
 ;
@@ -77,10 +77,10 @@ define amdgpu_kernel void @update_dppi64_test(ptr addrspace(1) %arg, i64 %in1, i
 ; GFX11-LABEL: update_dppi64_test:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT:    v_dual_mov_b32 v3, s3 :: v_dual_and_b32 v0, 0x3ff, v0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
+; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    global_load_b64 v[0:1], v4, s[0:1]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_mov_b32_dpp v2, v0 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
@@ -106,11 +106,11 @@ define amdgpu_kernel void @update_dppf64_test(ptr addrspace(1) %arg, double %in1
 ; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GFX8-NEXT:    flat_load_dwordx2 v[2:3], v[0:1]
-; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s2
+; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    v_mov_b32_dpp v4, v2 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    flat_store_dwordx2 v[0:1], v[4:5]
 ; GFX8-NEXT:    s_endpgm
 ;
@@ -131,10 +131,10 @@ define amdgpu_kernel void @update_dppf64_test(ptr addrspace(1) %arg, double %in1
 ; GFX11-LABEL: update_dppf64_test:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT:    v_dual_mov_b32 v3, s3 :: v_dual_and_b32 v0, 0x3ff, v0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
+; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    global_load_b64 v[0:1], v4, s[0:1]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_mov_b32_dpp v2, v0 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
@@ -160,11 +160,11 @@ define amdgpu_kernel void @update_dppv2i32_test(ptr addrspace(1) %arg, <2 x i32>
 ; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GFX8-NEXT:    flat_load_dwordx2 v[2:3], v[0:1]
-; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s2
+; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    v_mov_b32_dpp v4, v2 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    flat_store_dwordx2 v[0:1], v[4:5]
 ; GFX8-NEXT:    s_endpgm
 ;
@@ -185,10 +185,10 @@ define amdgpu_kernel void @update_dppv2i32_test(ptr addrspace(1) %arg, <2 x i32>
 ; GFX11-LABEL: update_dppv2i32_test:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT:    v_dual_mov_b32 v3, s3 :: v_dual_and_b32 v0, 0x3ff, v0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
+; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    global_load_b64 v[0:1], v4, s[0:1]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_mov_b32_dpp v2, v0 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
@@ -214,11 +214,11 @@ define amdgpu_kernel void @update_dppv2f32_test(ptr addrspace(1) %arg, <2 x floa
 ; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GFX8-NEXT:    flat_load_dwordx2 v[2:3], v[0:1]
-; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s2
+; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    v_mov_b32_dpp v4, v2 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    flat_store_dwordx2 v[0:1], v[4:5]
 ; GFX8-NEXT:    s_endpgm
 ;
@@ -239,10 +239,10 @@ define amdgpu_kernel void @update_dppv2f32_test(ptr addrspace(1) %arg, <2 x floa
 ; GFX11-LABEL: update_dppv2f32_test:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT:    v_dual_mov_b32 v3, s3 :: v_dual_and_b32 v0, 0x3ff, v0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
+; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    global_load_b64 v[0:1], v4, s[0:1]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_mov_b32_dpp v2, v0 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
@@ -268,11 +268,11 @@ define amdgpu_kernel void @update_dpp_p0_test(ptr addrspace(1) %arg, ptr %in1, p
 ; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GFX8-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GFX8-NEXT:    flat_load_dwordx2 v[2:3], v[0:1]
-; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    v_mov_b32_e32 v4, s2
+; GFX8-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    v_mov_b32_dpp v4, v2 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GFX8-NEXT:    v_mov_b32_dpp v5, v3 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
 ; GFX8-NEXT:    flat_store_dwordx2 v[0:1], v[4:5]
 ; GFX8-NEXT:    s_endpgm
 ;
@@ -293,10 +293,10 @@ define amdgpu_kernel void @update_dpp_p0_test(ptr addrspace(1) %arg, ptr %in1, p
 ; GFX11-LABEL: update_dpp_p0_test:
 ; GFX11:       ; %bb.0:
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX11-NEXT:    v_dual_mov_b32 v3, s3 :: v_dual_and_b32 v0, 0x3ff, v0
 ; GFX11-NEXT:    v_lshlrev_b32_e32 v4, 3, v0
+; GFX11-NEXT:    v_mov_b32_e32 v2, s2
 ; GFX11-NEXT:    global_load_b64 v[0:1], v4, s[0:1]
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_mov_b32_dpp v2, v0 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
index 495bfec5454ee8..98d2e71f3975c2 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
@@ -2819,343 +2819,327 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX8_DPP-LABEL: add_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
 ; GFX8_DPP-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8_DPP-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
+; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[4:5]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[4:5]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v4, vcc, v2, v4, vcc
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[4:5]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[4:5]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    v_readlane_b32 s7, v1, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s6, v4, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s7, v4, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s6, v3, 63
 ; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[4:5]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GFX8_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB5_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s7
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s6
 ; GFX8_DPP-NEXT:    s_mov_b32 s11, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s10, -1
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:    s_mov_b32 s8, s2
 ; GFX8_DPP-NEXT:    s_mov_b32 s9, s3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s6
-; GFX8_DPP-NEXT:    buffer_atomic_add_x2 v[7:8], off, s[8:11], 0 glc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s7
+; GFX8_DPP-NEXT:    buffer_atomic_add_x2 v[6:7], off, s[8:11], 0 glc
 ; GFX8_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8_DPP-NEXT:    buffer_wbinvl1_vol
 ; GFX8_DPP-NEXT:  .LBB5_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v7
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v7
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v6
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v1
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v3
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
-; GFX8_DPP-NEXT:    v_add_u32_e32 v7, vcc, s5, v7
+; GFX8_DPP-NEXT:    v_add_u32_e32 v6, vcc, s5, v6
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v8, vcc, v0, v8, vcc
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v7, vcc, v0, v7, vcc
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[6:7], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: add_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
 ; GFX9_DPP-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
+; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v4, vcc, v2, v4, vcc
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    v_readlane_b32 s3, v1, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s2, v4, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s3, v4, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GFX9_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB5_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s3
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s3
 ; GFX9_DPP-NEXT:    s_mov_b32 s11, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s10, -1
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:    s_mov_b32 s8, s6
 ; GFX9_DPP-NEXT:    s_mov_b32 s9, s7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s2
-; GFX9_DPP-NEXT:    buffer_atomic_add_x2 v[7:8], off, s[8:11], 0 glc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s2
+; GFX9_DPP-NEXT:    buffer_atomic_add_x2 v[6:7], off, s[8:11], 0 glc
 ; GFX9_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9_DPP-NEXT:    buffer_wbinvl1_vol
 ; GFX9_DPP-NEXT:  .LBB5_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v7
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v6
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v1
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v3
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v7, vcc, s1, v7
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v6, vcc, s1, v6
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v8, vcc, v0, v8, vcc
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v7, vcc, v0, v7, vcc
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[6:7], off, s[4:7], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: add_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v4
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v5, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v2, v1
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v3, v4, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v4
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v3, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x24
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v4, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v3, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s10, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s2, 16
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s3, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v3, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s11, v4, 47
-; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v4, 63
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s8, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s9, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v2, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v1, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s10, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s2, 16
+; GFX1064_DPP-NEXT:    v_writelane_b32 v6, s3, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v1, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s11, v2, 47
+; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v2, 63
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s8, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v6, s9, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s11, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s10, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s11, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v6, s10, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX1064_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[8:9], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB5_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s0
 ; GFX1064_DPP-NEXT:    s_mov_b32 s3, 0x31016000
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    s_mov_b32 s0, s6
 ; GFX1064_DPP-NEXT:    s_mov_b32 s1, s7
-; GFX1064_DPP-NEXT:    buffer_atomic_add_x2 v[9:10], off, s[0:3], 0 glc
+; GFX1064_DPP-NEXT:    buffer_atomic_add_x2 v[8:9], off, s[0:3], 0 glc
 ; GFX1064_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl1_inv
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB5_2:
 ; GFX1064_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[8:9]
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v12, v2
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v8
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v9
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1064_DPP-NEXT:    v_add_co_u32 v9, vcc, s0, v11
+; GFX1064_DPP-NEXT:    v_add_co_u32 v8, vcc, s0, v10
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, s2
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v10, vcc, s1, v12, vcc
-; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v9, vcc, s1, v11, vcc
+; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[8:9], off, s[4:7], 0
 ; GFX1064_DPP-NEXT:    s_endpgm
 ;
 ; GFX1032_DPP-LABEL: add_i64_varying:
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s0
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v4
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v5, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v2, v1
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v3, v4, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v4
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v3, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032_DPP-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x24
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s2, -1
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v7, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s8, v4, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s3, v3, 15
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v3
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s8, v2, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s3, v1, 15
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s2
 ; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s2, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s8, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s3, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v8, s8, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v7, s3, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s2
 ; GFX1032_DPP-NEXT:    s_mov_b32 s2, -1
 ; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
@@ -3177,8 +3161,8 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1032_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v8
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
@@ -3190,171 +3174,164 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ;
 ; GFX1164_DPP-LABEL: add_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 31
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s5
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v2, s4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s6, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s7, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s10, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s11, v3, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s8, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s9, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s6, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s7, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s10, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s11, v1, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s8, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s9, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[4:5], s[6:7]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s10, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s11, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s10, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s11, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[8:9], exec
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX1164_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB5_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s5
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s5
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s4
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    s_mov_b32 s4, s2
 ; GFX1164_DPP-NEXT:    s_mov_b32 s5, s3
-; GFX1164_DPP-NEXT:    buffer_atomic_add_u64 v[8:9], off, s[4:7], 0 glc
+; GFX1164_DPP-NEXT:    buffer_atomic_add_u64 v[6:7], off, s[4:7], 0 glc
 ; GFX1164_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl1_inv
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB5_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v11, v2
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s3, v9
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s2, v6
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v5
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s3, v7
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_u32 v8, vcc, s2, v10
+; GFX1164_DPP-NEXT:    v_add_co_u32 v6, vcc, s2, v8
 ; GFX1164_DPP-NEXT:    s_mov_b32 s2, s6
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v9, vcc, s3, v11, vcc
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v7, vcc, s3, v9, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b32 s3, 0x31016000
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[8:9], off, s[0:3], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[6:7], off, s[0:3], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: add_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v7, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s6, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
-; GFX1132_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s7, v4, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s8, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX1132_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s7, v2, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s8, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s6, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s7, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s8, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v6, s7, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v7, s8, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1132_DPP-NEXT:    s_mov_b32 s8, exec_lo
@@ -3375,8 +3352,8 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s3, v9
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1132_DPP-NEXT:    v_add_co_u32 v8, vcc_lo, s2, v10
@@ -3388,173 +3365,166 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ;
 ; GFX1264_DPP-LABEL: add_i64_varying:
 ; GFX1264_DPP:       ; %bb.0: ; %entry
-; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1264_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, v2
+; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1264_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v7, s5
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1264_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
+; GFX1264_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1264_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1264_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v2, s4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1264_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1264_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v4, 15
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v3, 15
-; GFX1264_DPP-NEXT:    v_readlane_b32 s8, v4, 31
-; GFX1264_DPP-NEXT:    v_readlane_b32 s9, v3, 31
-; GFX1264_DPP-NEXT:    v_writelane_b32 v1, s6, 16
-; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v4, 63
-; GFX1264_DPP-NEXT:    v_writelane_b32 v2, s7, 16
-; GFX1264_DPP-NEXT:    v_readlane_b32 s10, v4, 47
-; GFX1264_DPP-NEXT:    v_readlane_b32 s11, v3, 47
-; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v3, 63
-; GFX1264_DPP-NEXT:    v_writelane_b32 v1, s8, 32
-; GFX1264_DPP-NEXT:    v_writelane_b32 v2, s9, 32
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v2, 15
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v1, 15
+; GFX1264_DPP-NEXT:    v_readlane_b32 s8, v2, 31
+; GFX1264_DPP-NEXT:    v_readlane_b32 s9, v1, 31
+; GFX1264_DPP-NEXT:    v_writelane_b32 v4, s6, 16
+; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v2, 63
+; GFX1264_DPP-NEXT:    v_writelane_b32 v5, s7, 16
+; GFX1264_DPP-NEXT:    v_readlane_b32 s10, v2, 47
+; GFX1264_DPP-NEXT:    v_readlane_b32 s11, v1, 47
+; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v1, 63
+; GFX1264_DPP-NEXT:    v_writelane_b32 v4, s8, 32
+; GFX1264_DPP-NEXT:    v_writelane_b32 v5, s9, 32
 ; GFX1264_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1264_DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
 ; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
 ; GFX1264_DPP-NEXT:    s_mov_b64 s[4:5], s[6:7]
-; GFX1264_DPP-NEXT:    v_writelane_b32 v1, s10, 48
-; GFX1264_DPP-NEXT:    v_writelane_b32 v2, s11, 48
+; GFX1264_DPP-NEXT:    v_writelane_b32 v4, s10, 48
+; GFX1264_DPP-NEXT:    v_writelane_b32 v5, s11, 48
 ; GFX1264_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX1264_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1264_DPP-NEXT:    s_mov_b64 s[8:9], exec
-; GFX1264_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
+; GFX1264_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX1264_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
 ; GFX1264_DPP-NEXT:    s_cbranch_execz .LBB5_2
 ; GFX1264_DPP-NEXT:  ; %bb.1:
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v9, s5
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v8, s4
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v7, s5
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, s4
 ; GFX1264_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1264_DPP-NEXT:    s_wait_kmcnt 0x0
 ; GFX1264_DPP-NEXT:    s_mov_b32 s4, s2
 ; GFX1264_DPP-NEXT:    s_mov_b32 s5, s3
-; GFX1264_DPP-NEXT:    buffer_atomic_add_u64 v[8:9], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1264_DPP-NEXT:    buffer_atomic_add_u64 v[6:7], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX1264_DPP-NEXT:    s_wait_loadcnt 0x0
 ; GFX1264_DPP-NEXT:    global_inv scope:SCOPE_DEV
 ; GFX1264_DPP-NEXT:  .LBB5_2:
 ; GFX1264_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1264_DPP-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX1264_DPP-NEXT:    s_wait_kmcnt 0x0
-; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v11, v2
-; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s3, v9
+; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s2, v6
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v8, v4
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v9, v5
+; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s3, v7
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_add_co_u32 v8, vcc, s2, v10
+; GFX1264_DPP-NEXT:    v_add_co_u32 v6, vcc, s2, v8
 ; GFX1264_DPP-NEXT:    s_mov_b32 s2, s6
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v9, vcc, s3, v11, vcc
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v7, vcc, s3, v9, vcc
 ; GFX1264_DPP-NEXT:    s_mov_b32 s3, 0x31016000
-; GFX1264_DPP-NEXT:    buffer_store_b64 v[8:9], off, s[0:3], null
+; GFX1264_DPP-NEXT:    buffer_store_b64 v[6:7], off, s[0:3], null
 ; GFX1264_DPP-NEXT:    s_endpgm
 ;
 ; GFX1232_DPP-LABEL: add_i64_varying:
 ; GFX1232_DPP:       ; %bb.0: ; %entry
-; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1232_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s0
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
+; GFX1232_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v7, 0
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1232_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1232_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s6, -1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
-; GFX1232_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_readlane_b32 s7, v4, 15
-; GFX1232_DPP-NEXT:    v_readlane_b32 s8, v3, 15
-; GFX1232_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX1232_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_readlane_b32 s7, v2, 15
+; GFX1232_DPP-NEXT:    v_readlane_b32 s8, v1, 15
+; GFX1232_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1232_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s6, -1
-; GFX1232_DPP-NEXT:    v_writelane_b32 v1, s7, 16
-; GFX1232_DPP-NEXT:    v_writelane_b32 v2, s8, 16
+; GFX1232_DPP-NEXT:    v_writelane_b32 v6, s7, 16
+; GFX1232_DPP-NEXT:    v_writelane_b32 v7, s8, 16
 ; GFX1232_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1232_DPP-NEXT:    s_mov_b32 s6, -1
@@ -3576,8 +3546,8 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1232_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1232_DPP-NEXT:    s_wait_kmcnt 0x0
 ; GFX1232_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v11, v2
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX1232_DPP-NEXT:    v_readfirstlane_b32 s3, v9
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1232_DPP-NEXT:    v_add_co_u32 v8, vcc_lo, s2, v10
@@ -6448,343 +6418,327 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX8_DPP-LABEL: sub_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
 ; GFX8_DPP-NEXT:    s_load_dwordx4 s[0:3], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8_DPP-NEXT:    s_mov_b64 exec, s[4:5]
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
+; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[4:5]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[4:5]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v3, vcc, v3, v5
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v4, vcc, v2, v4, vcc
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[4:5]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[4:5]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    v_readlane_b32 s7, v1, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s6, v4, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s7, v4, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s6, v3, 63
 ; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[4:5]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GFX8_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB11_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s7
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s6
 ; GFX8_DPP-NEXT:    s_mov_b32 s11, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s10, -1
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:    s_mov_b32 s8, s2
 ; GFX8_DPP-NEXT:    s_mov_b32 s9, s3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s6
-; GFX8_DPP-NEXT:    buffer_atomic_sub_x2 v[7:8], off, s[8:11], 0 glc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s7
+; GFX8_DPP-NEXT:    buffer_atomic_sub_x2 v[6:7], off, s[8:11], 0 glc
 ; GFX8_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8_DPP-NEXT:    buffer_wbinvl1_vol
 ; GFX8_DPP-NEXT:  .LBB11_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v7
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v7
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v6
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v1
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v3
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
-; GFX8_DPP-NEXT:    v_sub_u32_e32 v7, vcc, s5, v7
+; GFX8_DPP-NEXT:    v_sub_u32_e32 v6, vcc, s5, v6
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX8_DPP-NEXT:    v_subb_u32_e32 v8, vcc, v0, v8, vcc
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX8_DPP-NEXT:    v_subb_u32_e32 v7, vcc, v0, v7, vcc
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[6:7], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: sub_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
 ; GFX9_DPP-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
+; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v2, vcc, v2, v4, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v4, vcc, v2, v4, vcc
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    v_readlane_b32 s3, v1, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s2, v4, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s3, v4, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GFX9_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB11_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s3
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s3
 ; GFX9_DPP-NEXT:    s_mov_b32 s11, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s10, -1
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:    s_mov_b32 s8, s6
 ; GFX9_DPP-NEXT:    s_mov_b32 s9, s7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s2
-; GFX9_DPP-NEXT:    buffer_atomic_sub_x2 v[7:8], off, s[8:11], 0 glc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s2
+; GFX9_DPP-NEXT:    buffer_atomic_sub_x2 v[6:7], off, s[8:11], 0 glc
 ; GFX9_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9_DPP-NEXT:    buffer_wbinvl1_vol
 ; GFX9_DPP-NEXT:  .LBB11_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v7
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v6
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v1
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v3
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9_DPP-NEXT:    v_sub_co_u32_e32 v7, vcc, s1, v7
+; GFX9_DPP-NEXT:    v_sub_co_u32_e32 v6, vcc, s1, v6
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX9_DPP-NEXT:    v_subb_co_u32_e32 v8, vcc, v0, v8, vcc
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
+; GFX9_DPP-NEXT:    v_subb_co_u32_e32 v7, vcc, v0, v7, vcc
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[6:7], off, s[4:7], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: sub_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v4
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v5, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v2, v1
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v3, v4, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v4
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v3, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x24
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v4, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v3, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s10, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s2, 16
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s3, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v3, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s11, v4, 47
-; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v4, 63
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s8, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s9, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v2, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v1, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s10, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s2, 16
+; GFX1064_DPP-NEXT:    v_writelane_b32 v6, s3, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s2, v1, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s11, v2, 47
+; GFX1064_DPP-NEXT:    v_readlane_b32 s3, v2, 63
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s8, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v6, s9, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s11, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s10, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s11, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v6, s10, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX1064_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[8:9], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB11_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s0
 ; GFX1064_DPP-NEXT:    s_mov_b32 s3, 0x31016000
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    s_mov_b32 s0, s6
 ; GFX1064_DPP-NEXT:    s_mov_b32 s1, s7
-; GFX1064_DPP-NEXT:    buffer_atomic_sub_x2 v[9:10], off, s[0:3], 0 glc
+; GFX1064_DPP-NEXT:    buffer_atomic_sub_x2 v[8:9], off, s[0:3], 0 glc
 ; GFX1064_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl1_inv
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB11_2:
 ; GFX1064_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[8:9]
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v12, v2
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v8
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v9
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1064_DPP-NEXT:    v_sub_co_u32 v9, vcc, s0, v11
+; GFX1064_DPP-NEXT:    v_sub_co_u32 v8, vcc, s0, v10
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, s2
-; GFX1064_DPP-NEXT:    v_sub_co_ci_u32_e32 v10, vcc, s1, v12, vcc
-; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1064_DPP-NEXT:    v_sub_co_ci_u32_e32 v9, vcc, s1, v11, vcc
+; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[8:9], off, s[4:7], 0
 ; GFX1064_DPP-NEXT:    s_endpgm
 ;
 ; GFX1032_DPP-LABEL: sub_i64_varying:
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s0
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v4
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v5, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v2, v1
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v3, v4, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v4
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v3, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032_DPP-NEXT:    s_load_dwordx4 s[4:7], s[2:3], 0x24
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s2, -1
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v7, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s8, v4, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s3, v3, 15
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v3
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s8, v2, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s3, v1, 15
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s2
 ; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s2, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s8, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s3, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v8, s8, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v7, s3, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s2
 ; GFX1032_DPP-NEXT:    s_mov_b32 s2, -1
 ; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
@@ -6806,8 +6760,8 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1032_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v8
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
@@ -6819,171 +6773,164 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ;
 ; GFX1164_DPP-LABEL: sub_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 31
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s5
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v2, s4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s6, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s7, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s10, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s11, v3, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s8, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s9, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s6, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s7, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s10, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s11, v1, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s8, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s9, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[4:5], s[6:7]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s10, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s11, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s10, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s11, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[8:9], exec
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX1164_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB11_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s5
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s5
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s4
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    s_mov_b32 s4, s2
 ; GFX1164_DPP-NEXT:    s_mov_b32 s5, s3
-; GFX1164_DPP-NEXT:    buffer_atomic_sub_u64 v[8:9], off, s[4:7], 0 glc
+; GFX1164_DPP-NEXT:    buffer_atomic_sub_u64 v[6:7], off, s[4:7], 0 glc
 ; GFX1164_DPP-NEXT:    s_waitcnt vmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl1_inv
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB11_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v11, v2
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s3, v9
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s2, v6
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v5
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s3, v7
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_sub_co_u32 v8, vcc, s2, v10
+; GFX1164_DPP-NEXT:    v_sub_co_u32 v6, vcc, s2, v8
 ; GFX1164_DPP-NEXT:    s_mov_b32 s2, s6
-; GFX1164_DPP-NEXT:    v_sub_co_ci_u32_e32 v9, vcc, s3, v11, vcc
+; GFX1164_DPP-NEXT:    v_sub_co_ci_u32_e32 v7, vcc, s3, v9, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b32 s3, 0x31016000
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[8:9], off, s[0:3], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[6:7], off, s[0:3], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: sub_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v7, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s6, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
-; GFX1132_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s7, v4, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s8, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX1132_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s7, v2, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s8, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s6, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s7, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s8, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v6, s7, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v7, s8, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1132_DPP-NEXT:    s_mov_b32 s8, exec_lo
@@ -7004,8 +6951,8 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s3, v9
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1132_DPP-NEXT:    v_sub_co_u32 v8, vcc_lo, s2, v10
@@ -7017,173 +6964,166 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ;
 ; GFX1264_DPP-LABEL: sub_i64_varying:
 ; GFX1264_DPP:       ; %bb.0: ; %entry
-; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1264_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, v2
+; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1264_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1264_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v7, s5
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1264_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
+; GFX1264_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1264_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1264_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1264_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v2, s4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1264_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1264_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1264_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v4, 15
-; GFX1264_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v3, 15
-; GFX1264_DPP-NEXT:    v_readlane_b32 s8, v4, 31
-; GFX1264_DPP-NEXT:    v_readlane_b32 s9, v3, 31
-; GFX1264_DPP-NEXT:    v_writelane_b32 v1, s6, 16
-; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v4, 63
-; GFX1264_DPP-NEXT:    v_writelane_b32 v2, s7, 16
-; GFX1264_DPP-NEXT:    v_readlane_b32 s10, v4, 47
-; GFX1264_DPP-NEXT:    v_readlane_b32 s11, v3, 47
-; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v3, 63
-; GFX1264_DPP-NEXT:    v_writelane_b32 v1, s8, 32
-; GFX1264_DPP-NEXT:    v_writelane_b32 v2, s9, 32
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v2, 15
+; GFX1264_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v1, 15
+; GFX1264_DPP-NEXT:    v_readlane_b32 s8, v2, 31
+; GFX1264_DPP-NEXT:    v_readlane_b32 s9, v1, 31
+; GFX1264_DPP-NEXT:    v_writelane_b32 v4, s6, 16
+; GFX1264_DPP-NEXT:    v_readlane_b32 s6, v2, 63
+; GFX1264_DPP-NEXT:    v_writelane_b32 v5, s7, 16
+; GFX1264_DPP-NEXT:    v_readlane_b32 s10, v2, 47
+; GFX1264_DPP-NEXT:    v_readlane_b32 s11, v1, 47
+; GFX1264_DPP-NEXT:    v_readlane_b32 s7, v1, 63
+; GFX1264_DPP-NEXT:    v_writelane_b32 v4, s8, 32
+; GFX1264_DPP-NEXT:    v_writelane_b32 v5, s9, 32
 ; GFX1264_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[4:5]
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1264_DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
 ; GFX1264_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
 ; GFX1264_DPP-NEXT:    s_mov_b64 s[4:5], s[6:7]
-; GFX1264_DPP-NEXT:    v_writelane_b32 v1, s10, 48
-; GFX1264_DPP-NEXT:    v_writelane_b32 v2, s11, 48
+; GFX1264_DPP-NEXT:    v_writelane_b32 v4, s10, 48
+; GFX1264_DPP-NEXT:    v_writelane_b32 v5, s11, 48
 ; GFX1264_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1264_DPP-NEXT:    s_mov_b64 exec, s[8:9]
 ; GFX1264_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1264_DPP-NEXT:    s_mov_b64 s[8:9], exec
-; GFX1264_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
+; GFX1264_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX1264_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
 ; GFX1264_DPP-NEXT:    s_cbranch_execz .LBB11_2
 ; GFX1264_DPP-NEXT:  ; %bb.1:
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v9, s5
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v8, s4
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v7, s5
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v6, s4
 ; GFX1264_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1264_DPP-NEXT:    s_wait_kmcnt 0x0
 ; GFX1264_DPP-NEXT:    s_mov_b32 s4, s2
 ; GFX1264_DPP-NEXT:    s_mov_b32 s5, s3
-; GFX1264_DPP-NEXT:    buffer_atomic_sub_u64 v[8:9], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1264_DPP-NEXT:    buffer_atomic_sub_u64 v[6:7], off, s[4:7], null th:TH_ATOMIC_RETURN scope:SCOPE_DEV
 ; GFX1264_DPP-NEXT:    s_wait_loadcnt 0x0
 ; GFX1264_DPP-NEXT:    global_inv scope:SCOPE_DEV
 ; GFX1264_DPP-NEXT:  .LBB11_2:
 ; GFX1264_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1264_DPP-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GFX1264_DPP-NEXT:    s_wait_kmcnt 0x0
-; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1264_DPP-NEXT:    v_mov_b32_e32 v11, v2
-; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s3, v9
+; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s2, v6
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v8, v4
+; GFX1264_DPP-NEXT:    v_mov_b32_e32 v9, v5
+; GFX1264_DPP-NEXT:    v_readfirstlane_b32 s3, v7
 ; GFX1264_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1264_DPP-NEXT:    v_sub_co_u32 v8, vcc, s2, v10
+; GFX1264_DPP-NEXT:    v_sub_co_u32 v6, vcc, s2, v8
 ; GFX1264_DPP-NEXT:    s_mov_b32 s2, s6
-; GFX1264_DPP-NEXT:    v_sub_co_ci_u32_e32 v9, vcc, s3, v11, vcc
+; GFX1264_DPP-NEXT:    v_sub_co_ci_u32_e32 v7, vcc, s3, v9, vcc
 ; GFX1264_DPP-NEXT:    s_mov_b32 s3, 0x31016000
-; GFX1264_DPP-NEXT:    buffer_store_b64 v[8:9], off, s[0:3], null
+; GFX1264_DPP-NEXT:    buffer_store_b64 v[6:7], off, s[0:3], null
 ; GFX1264_DPP-NEXT:    s_endpgm
 ;
 ; GFX1232_DPP-LABEL: sub_i64_varying:
 ; GFX1232_DPP:       ; %bb.0: ; %entry
-; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1232_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s0
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1232_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
+; GFX1232_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v7, 0
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
-; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1232_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1232_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1232_DPP-NEXT:    s_load_b128 s[0:3], s[2:3], 0x24
 ; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s6, -1
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
-; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
-; GFX1232_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1232_DPP-NEXT:    v_readlane_b32 s7, v4, 15
-; GFX1232_DPP-NEXT:    v_readlane_b32 s8, v3, 15
-; GFX1232_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1232_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
+; GFX1232_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1232_DPP-NEXT:    v_readlane_b32 s7, v2, 15
+; GFX1232_DPP-NEXT:    v_readlane_b32 s8, v1, 15
+; GFX1232_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1232_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1232_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1232_DPP-NEXT:    s_or_saveexec_b32 s6, -1
-; GFX1232_DPP-NEXT:    v_writelane_b32 v1, s7, 16
-; GFX1232_DPP-NEXT:    v_writelane_b32 v2, s8, 16
+; GFX1232_DPP-NEXT:    v_writelane_b32 v6, s7, 16
+; GFX1232_DPP-NEXT:    v_writelane_b32 v7, s8, 16
 ; GFX1232_DPP-NEXT:    s_wait_alu 0xfffe
 ; GFX1232_DPP-NEXT:    s_mov_b32 exec_lo, s6
 ; GFX1232_DPP-NEXT:    s_mov_b32 s6, -1
@@ -7205,8 +7145,8 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1232_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1232_DPP-NEXT:    s_wait_kmcnt 0x0
 ; GFX1232_DPP-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1232_DPP-NEXT:    v_mov_b32_e32 v11, v2
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1232_DPP-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX1232_DPP-NEXT:    v_readfirstlane_b32 s3, v9
 ; GFX1232_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
 ; GFX1232_DPP-NEXT:    v_sub_co_u32 v8, vcc_lo, s2, v10
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
index 5924649ebe7350..f7773b4859dc2c 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll
@@ -2333,250 +2333,230 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX8_DPP-LABEL: add_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v5, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    v_readlane_b32 s5, v1, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s4, v4, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_readlane_b32 s4, v2, 63
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
+; GFX8_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB6_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, s4
 ; GFX8_DPP-NEXT:    s_mov_b32 m0, -1
-; GFX8_DPP-NEXT:    ds_add_rtn_u64 v[7:8], v9, v[7:8]
+; GFX8_DPP-NEXT:    ds_add_rtn_u64 v[5:6], v7, v[5:6]
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB6_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
 ; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v3
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
-; GFX8_DPP-NEXT:    v_add_u32_e32 v7, vcc, s5, v7
+; GFX8_DPP-NEXT:    v_add_u32_e32 v5, vcc, s5, v5
 ; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v8, vcc, v0, v8, vcc
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v6, vcc, v0, v6, vcc
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: add_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v5, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    v_readlane_b32 s5, v1, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s4, v4, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_readlane_b32 s4, v2, 63
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
+; GFX9_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB6_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX9_DPP-NEXT:    ds_add_rtn_u64 v[7:8], v9, v[7:8]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX9_DPP-NEXT:    ds_add_rtn_u64 v[5:6], v7, v[5:6]
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB6_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
 ; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v3
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s4
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v7, vcc, s5, v7
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v5, vcc, s5, v5
 ; GFX9_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v8, vcc, v0, v8, vcc
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v6, vcc, v0, v6, vcc
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: add_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v4
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v5, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v2, v1
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v3, v4, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v4
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v3, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v4, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v8, s4, 16
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s5, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v2, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v8, s6, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s7, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s9, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s8, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v8, s9, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s8, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
@@ -2594,8 +2574,8 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1064_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
 ; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v12, v2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v12, v8
 ; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1064_DPP-NEXT:    v_add_co_u32 v9, vcc, s0, v11
@@ -2608,53 +2588,53 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s4
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v4
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v5, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v2, v1
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v3, v4, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v4
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v3, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v7, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v4, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v3
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v2, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v1, 15
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s6, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s5, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v8, s6, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v7, s5, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
@@ -2672,8 +2652,8 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v8
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1032_DPP-NEXT:    v_add_co_u32 v9, vcc_lo, s0, v11
@@ -2684,165 +2664,158 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX1164_DPP-LABEL: add_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s5
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v2, s4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s4, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 63
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s6, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s7, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v6, s5, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 63
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v6, s7, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s8, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s9, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v6, s9, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v8
+; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1164_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB6_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s0
-; GFX1164_DPP-NEXT:    ds_add_rtn_u64 v[8:9], v0, v[8:9]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1164_DPP-NEXT:    ds_add_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB6_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v8
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v11, v2
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v9
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v5
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v8
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_u32 v8, vcc, s0, v10
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v9, vcc, s1, v11, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32 v7, vcc, s0, v9
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v8, vcc, s1, v10, vcc
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[8:9], off, s[4:7], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: add_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s4
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v7, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v4, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v4, 15
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v2, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v2, 15
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v1, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
 ; GFX1132_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s6, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v6, s5, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v7, s6, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v8
@@ -2858,8 +2831,8 @@ define amdgpu_kernel void @add_i64_varying(ptr addrspace(1) %out) {
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v8
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v9
 ; GFX1132_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
@@ -3125,137 +3098,119 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ;
 ; GFX8_DPP-LABEL: add_i64_varying_nouse:
 ; GFX8_DPP:       ; %bb.0: ; %entry
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, 0
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v8
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    s_nop 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
 ; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v4, v2
 ; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX8_DPP-NEXT:    v_readlane_b32 s3, v1, 63
 ; GFX8_DPP-NEXT:    v_readlane_b32 s2, v2, 63
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX8_DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB7_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, s1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s0
 ; GFX8_DPP-NEXT:    s_mov_b32 m0, -1
-; GFX8_DPP-NEXT:    ds_add_u64 v7, v[8:9]
+; GFX8_DPP-NEXT:    ds_add_u64 v5, v[6:7]
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB7_2:
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: add_i64_varying_nouse:
 ; GFX9_DPP:       ; %bb.0: ; %entry
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, 0
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v8
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    s_nop 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
 ; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v4, v2
 ; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX9_DPP-NEXT:    v_readlane_b32 s3, v1, 63
 ; GFX9_DPP-NEXT:    v_readlane_b32 s2, v2, 63
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9_DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB7_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, s1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s0
-; GFX9_DPP-NEXT:    ds_add_u64 v7, v[8:9]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s0
+; GFX9_DPP-NEXT:    ds_add_u64 v5, v[6:7]
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB7_2:
 ; GFX9_DPP-NEXT:    s_endpgm
@@ -3264,31 +3219,31 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v4
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v5, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v3, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v2, v1
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v3, v4, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v4
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v3, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
 ; GFX1064_DPP-NEXT:    v_permlanex16_b32 v3, v1, 0, 0
 ; GFX1064_DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
 ; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
@@ -3301,17 +3256,17 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 32
 ; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064_DPP-NEXT:    s_add_u32 s0, s3, s4
 ; GFX1064_DPP-NEXT:    s_addc_u32 s1, s2, s5
-; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB7_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1064_DPP-NEXT:    ds_add_u64 v0, v[9:10]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1064_DPP-NEXT:    ds_add_u64 v0, v[7:8]
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB7_2:
@@ -3321,45 +3276,45 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s0
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v4
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v5, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v3, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v2, v1
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v3, v4, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v4
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v3, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
 ; GFX1032_DPP-NEXT:    v_permlanex16_b32 v3, v1, 0, 0
 ; GFX1032_DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
 ; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v3
 ; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v4, vcc_lo
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v11, exec_lo, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
+; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, v2
-; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v11
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
 ; GFX1032_DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032_DPP-NEXT:    s_cbranch_execz .LBB7_2
 ; GFX1032_DPP-NEXT:  ; %bb.1:
-; GFX1032_DPP-NEXT:    ds_add_u64 v0, v[9:10]
+; GFX1032_DPP-NEXT:    ds_add_u64 v0, v[7:8]
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    buffer_gl0_inv
 ; GFX1032_DPP-NEXT:  .LBB7_2:
@@ -3367,44 +3322,41 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ;
 ; GFX1164_DPP-LABEL: add_i64_varying_nouse:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v4, v4 row_xmask:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_xmask:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v4, v4 row_xmask:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_xmask:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v5, vcc
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v4, v4 row_xmask:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v4, v4 row_xmask:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_xmask:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v2, v3, 0, 0
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v4, v1, 0, 0
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_xmask:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v3, v2, 0, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_u32 v2, vcc, v3, v2
 ; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32 v2, vcc, v2, v3
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v4, v1, 0, 0
 ; GFX1164_DPP-NEXT:    v_permlane64_b32 v3, v2
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1164_DPP-NEXT:    v_permlane64_b32 v4, v1
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
@@ -3413,16 +3365,16 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ; GFX1164_DPP-NEXT:    v_add_co_u32 v2, vcc, v2, v3
 ; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v1, v4, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v3
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v7
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB7_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    ds_add_u64 v0, v[6:7]
+; GFX1164_DPP-NEXT:    ds_add_u64 v0, v[5:6]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB7_2:
@@ -3430,53 +3382,50 @@ define amdgpu_kernel void @add_i64_varying_nouse() {
 ;
 ; GFX1132_DPP-LABEL: add_i64_varying_nouse:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v4, v4 row_xmask:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_xmask:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v4, v4 row_xmask:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_xmask:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v5, vcc_lo
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v4, v4 row_xmask:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v4, v4 row_xmask:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_xmask:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v2, v3, 0, 0
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v4, v1, 0, 0
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_xmask:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_u32 v2, vcc_lo, v3, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v3, v2, 0, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32 v2, vcc_lo, v2, v3
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v4, v1, 0, 0
 ; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v1, v4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v7, v3
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v5, v2
+; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v3
 ; GFX1132_DPP-NEXT:    s_mov_b32 s0, exec_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132_DPP-NEXT:    v_cmpx_eq_u32_e32 0, v7
 ; GFX1132_DPP-NEXT:    s_cbranch_execz .LBB7_2
 ; GFX1132_DPP-NEXT:  ; %bb.1:
-; GFX1132_DPP-NEXT:    ds_add_u64 v0, v[6:7]
+; GFX1132_DPP-NEXT:    ds_add_u64 v0, v[5:6]
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    buffer_gl0_inv
 ; GFX1132_DPP-NEXT:  .LBB7_2:
@@ -5835,250 +5784,230 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX8_DPP-LABEL: sub_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v5, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX8_DPP-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_addc_u32_e32 v1, vcc, v1, v6, vcc
 ; GFX8_DPP-NEXT:    v_readlane_b32 s5, v1, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s4, v4, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_readlane_b32 s4, v2, 63
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
+; GFX8_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB14_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, s4
 ; GFX8_DPP-NEXT:    s_mov_b32 m0, -1
-; GFX8_DPP-NEXT:    ds_sub_rtn_u64 v[7:8], v9, v[7:8]
+; GFX8_DPP-NEXT:    ds_sub_rtn_u64 v[5:6], v7, v[5:6]
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB14_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
 ; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v3
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
-; GFX8_DPP-NEXT:    v_sub_u32_e32 v7, vcc, s5, v7
+; GFX8_DPP-NEXT:    v_sub_u32_e32 v5, vcc, s5, v5
 ; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX8_DPP-NEXT:    v_subb_u32_e32 v8, vcc, v0, v8, vcc
+; GFX8_DPP-NEXT:    v_subb_u32_e32 v6, vcc, v0, v6, vcc
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: sub_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v5, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v3
-; GFX9_DPP-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v5
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v6, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    v_readlane_b32 s5, v1, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s4, v4, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_readlane_b32 s4, v2, 63
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
+; GFX9_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB14_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX9_DPP-NEXT:    ds_sub_rtn_u64 v[7:8], v9, v[7:8]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX9_DPP-NEXT:    ds_sub_rtn_u64 v[5:6], v7, v[5:6]
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB14_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
 ; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v3
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s4
-; GFX9_DPP-NEXT:    v_sub_co_u32_e32 v7, vcc, s5, v7
+; GFX9_DPP-NEXT:    v_sub_co_u32_e32 v5, vcc, s5, v5
 ; GFX9_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX9_DPP-NEXT:    v_subb_co_u32_e32 v8, vcc, v0, v8, vcc
+; GFX9_DPP-NEXT:    v_subb_co_u32_e32 v6, vcc, v0, v6, vcc
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: sub_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v4
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v5, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v6, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v7
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v8, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v2, v1
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v3, v4, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v4
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v3, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v6
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_add_co_u32 v3, vcc, v3, v5
-; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc, v4, v7, vcc
+; GFX1064_DPP-NEXT:    v_add_co_u32 v1, vcc, v1, v3
+; GFX1064_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc, v2, v5, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v4, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v8, s4, 16
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s5, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v2, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v8, s6, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s7, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s9, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s8, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v8, s9, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v7, s8, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
@@ -6096,8 +6025,8 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out) {
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1064_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
 ; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v12, v2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v12, v8
 ; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1064_DPP-NEXT:    v_sub_co_u32 v9, vcc, s0, v11
@@ -6110,53 +6039,53 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, v1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, 0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s4
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v5 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v4
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v5, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v6, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v7
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v2
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v8, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v4, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v2, v1
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v3, v4, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v4
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v3, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v6
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_add_co_u32 v3, vcc_lo, v3, v5
-; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v4, v7, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v4, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_add_co_u32 v1, vcc_lo, v1, v3
+; GFX1032_DPP-NEXT:    v_add_co_ci_u32_e32 v2, vcc_lo, v2, v5, vcc_lo
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v2, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v1, 15
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s6, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s5, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v8, s6, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v7, s5, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
@@ -6174,8 +6103,8 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v11, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v12, v8
 ; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1032_DPP-NEXT:    v_sub_co_u32 v9, vcc_lo, s0, v11
@@ -6186,165 +6115,158 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX1164_DPP-LABEL: sub_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v4, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v2, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v6, vcc
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 31
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s5
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v2
-; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc, v7, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v4, vcc
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s4
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc, v3, v5, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 31
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, s4
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc, v2, s4 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s4, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 63
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s6, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s7, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v6, s5, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 63
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v6, s7, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s8, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s9, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v6, s9, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v8
+; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr8_vgpr9
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1164_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB14_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s0
-; GFX1164_DPP-NEXT:    ds_sub_rtn_u64 v[8:9], v0, v[8:9]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1164_DPP-NEXT:    ds_sub_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB14_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v8
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v11, v2
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v9
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v5
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v8
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_sub_co_u32 v8, vcc, s0, v10
-; GFX1164_DPP-NEXT:    v_sub_co_ci_u32_e32 v9, vcc, s1, v11, vcc
+; GFX1164_DPP-NEXT:    v_sub_co_u32 v7, vcc, s0, v9
+; GFX1164_DPP-NEXT:    v_sub_co_ci_u32_e32 v8, vcc, s1, v10, vcc
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[8:9], off, s[4:7], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: sub_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, 0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s4
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v7, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v5, vcc_lo, v5, v5 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v3, vcc_lo, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v4, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v5, v5 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, v2
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v6, vcc_lo
-; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v4, vcc_lo, v7, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v4, vcc_lo
+; GFX1132_DPP-NEXT:    v_add_co_u32_e64_dpp v2, vcc_lo, v5, v2 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v3, -1, -1
-; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v4, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v4, 15
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v4, v1, -1, -1
+; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v2, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v2, 15
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, v3, v5, vcc_lo
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v4 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v1, v3, vcc_lo
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v1, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
 ; GFX1132_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s6, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v6, s5, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v7, s6, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
 ; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v8
@@ -6360,8 +6282,8 @@ define amdgpu_kernel void @sub_i64_varying(ptr addrspace(1) %out) {
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v8
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v6
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v11, v7
 ; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v9
 ; GFX1132_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
@@ -7339,30 +7261,30 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, 0, s[0:1]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, -1
-; GFX8_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX8_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
-; GFX8_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, 0, s[0:1]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, -1, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v1, -1
+; GFX8_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, -1
+; GFX8_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_readlane_b32 s5, v1, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s4, v2, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    s_nop 0
+; GFX8_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_readlane_b32 s5, v3, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s4, v4, 63
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v1, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
 ; GFX8_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
@@ -7376,16 +7298,17 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB16_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX8_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s0, v6
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s1, v5
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX8_DPP-NEXT:    s_mov_b32 s7, 0xf000
-; GFX8_DPP-NEXT:    v_and_b32_e32 v6, s0, v6
-; GFX8_DPP-NEXT:    v_and_b32_e32 v5, s1, v5
+; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v2
+; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
+; GFX8_DPP-NEXT:    v_and_b32_e32 v6, s4, v6
+; GFX8_DPP-NEXT:    v_and_b32_e32 v5, s5, v5
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[4:7], 0
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: and_i64_varying:
@@ -7394,30 +7317,30 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, -1
-; GFX9_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX9_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
-; GFX9_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, 0, s[0:1]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, -1, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v1, -1
+; GFX9_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, -1
+; GFX9_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_readlane_b32 s5, v1, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s4, v2, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    s_nop 0
+; GFX9_DPP-NEXT:    v_and_b32_dpp v3, v3, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_and_b32_dpp v4, v4, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_readlane_b32 s5, v3, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s4, v4, 63
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v1, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
 ; GFX9_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
@@ -7430,16 +7353,17 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB16_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX9_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v6
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v5
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX9_DPP-NEXT:    s_mov_b32 s7, 0xf000
-; GFX9_DPP-NEXT:    v_and_b32_e32 v6, s0, v6
-; GFX9_DPP-NEXT:    v_and_b32_e32 v5, s1, v5
+; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v2
+; GFX9_DPP-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9_DPP-NEXT:    s_mov_b32 s2, -1
+; GFX9_DPP-NEXT:    v_and_b32_e32 v6, s4, v6
+; GFX9_DPP-NEXT:    v_and_b32_e32 v5, s5, v5
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[4:7], 0
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: and_i64_varying:
@@ -7447,8 +7371,8 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, 0, s[0:1]
 ; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX1064_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1064_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf
@@ -7523,8 +7447,8 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, 0, s4
 ; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX1032_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf
@@ -7583,8 +7507,8 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, 0, s[0:1]
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1164_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf
@@ -7672,9 +7596,8 @@ define amdgpu_kernel void @and_i64_varying(ptr addrspace(1) %out) {
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, 0, s4
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, -1
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, -1 :: v_dual_mov_b32 v5, -1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1132_DPP-NEXT:    v_and_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    v_and_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
@@ -8702,10 +8625,10 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX8_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX8_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX8_DPP-NEXT:    s_nop 0
@@ -8722,8 +8645,8 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
 ; GFX8_DPP-NEXT:    v_readlane_b32 s5, v1, 63
 ; GFX8_DPP-NEXT:    v_readlane_b32 s4, v2, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
 ; GFX8_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
@@ -8758,10 +8681,10 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX9_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX9_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX9_DPP-NEXT:    s_nop 0
@@ -8778,8 +8701,8 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
 ; GFX9_DPP-NEXT:    v_readlane_b32 s5, v1, 63
 ; GFX9_DPP-NEXT:    v_readlane_b32 s4, v2, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
 ; GFX9_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
@@ -8810,8 +8733,8 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1064_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1064_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1064_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
@@ -8886,8 +8809,8 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s4
 ; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1032_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1032_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1032_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
@@ -8946,8 +8869,8 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1164_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
@@ -9035,9 +8958,8 @@ define amdgpu_kernel void @or_i64_varying(ptr addrspace(1) %out) {
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s4
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, 0
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v5, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1132_DPP-NEXT:    v_or_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    v_or_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
@@ -10065,10 +9987,10 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX8_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX8_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX8_DPP-NEXT:    s_nop 0
@@ -10085,8 +10007,8 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
 ; GFX8_DPP-NEXT:    v_readlane_b32 s5, v1, 63
 ; GFX8_DPP-NEXT:    v_readlane_b32 s4, v2, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
 ; GFX8_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
@@ -10121,10 +10043,10 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX9_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX9_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX9_DPP-NEXT:    s_nop 0
@@ -10141,8 +10063,8 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
 ; GFX9_DPP-NEXT:    v_readlane_b32 s5, v1, 63
 ; GFX9_DPP-NEXT:    v_readlane_b32 s4, v2, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
 ; GFX9_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
@@ -10173,8 +10095,8 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1064_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1064_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1064_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
@@ -10249,8 +10171,8 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s4
 ; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1032_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1032_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1032_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
@@ -10309,8 +10231,8 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s[0:1]
 ; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
 ; GFX1164_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1164_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
@@ -10398,9 +10320,8 @@ define amdgpu_kernel void @xor_i64_varying(ptr addrspace(1) %out) {
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, 0, s4
 ; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, v0, s4
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v6, 0
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v5, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX1132_DPP-NEXT:    v_xor_b32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    v_xor_b32_dpp v2, v2, v2 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
@@ -11711,71 +11632,66 @@ define amdgpu_kernel void @max_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX8_DPP-LABEL: max_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
-; GFX8_DPP-NEXT:    s_mov_b32 s0, 0
 ; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX8_DPP-NEXT:    s_brev_b32 s1, 1
 ; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
-; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, s1
-; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v1, 1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, s0
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, v1, 0, s[4:5]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[4:5]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v2, 1
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, v2, 0, s[0:1]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_readlane_b32 s1, v2, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s0, v1, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    s_mov_b64 exec, s[4:5]
-; GFX8_DPP-NEXT:    s_mov_b64 s[4:5], s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_readlane_b32 s5, v4, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s4, v3, 63
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
@@ -11790,9 +11706,9 @@ define amdgpu_kernel void @max_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
 ; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v8
 ; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v7
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v1
 ; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v4
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX8_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, s[4:5], v[7:8]
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
@@ -11806,71 +11722,66 @@ define amdgpu_kernel void @max_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX9_DPP-LABEL: max_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
-; GFX9_DPP-NEXT:    s_mov_b32 s0, 0
 ; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9_DPP-NEXT:    s_brev_b32 s1, 1
 ; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
-; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[4:5], -1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, s1
-; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v1, 1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, s0
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, v1, 0, s[4:5]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[4:5]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v2, 1
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, v2, 0, s[0:1]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_readlane_b32 s1, v2, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s0, v1, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    s_mov_b64 exec, s[4:5]
-; GFX9_DPP-NEXT:    s_mov_b64 s[4:5], s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_readlane_b32 s5, v4, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s4, v3, 63
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
@@ -11884,9 +11795,9 @@ define amdgpu_kernel void @max_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
 ; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v8
 ; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v1
 ; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v4
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX9_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, s[4:5], v[7:8]
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
@@ -11901,412 +11812,404 @@ define amdgpu_kernel void @max_i64_varying(ptr addrspace(1) %out) {
 ; GFX1064_DPP-LABEL: max_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    s_mov_b32 s4, 0
-; GFX1064_DPP-NEXT:    s_brev_b32 s5, 1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, s5
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x80000000, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x80000000, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s8, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s9, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s9, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB23_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1064_DPP-NEXT:    ds_max_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1064_DPP-NEXT:    ds_max_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB23_2:
 ; GFX1064_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1064_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, s[0:1], v[9:10]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1064_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, s[0:1], v[7:8]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1064_DPP-NEXT:    s_endpgm
 ;
 ; GFX1032_DPP-LABEL: max_i64_varying:
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    s_mov_b32 s0, 0
-; GFX1032_DPP-NEXT:    s_brev_b32 s1, 1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x80000000, 0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, s1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, s0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x80000000, 0, s4
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s4
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s6, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1032_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1032_DPP-NEXT:    s_cbranch_execz .LBB23_2
 ; GFX1032_DPP-NEXT:  ; %bb.1:
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1032_DPP-NEXT:    ds_max_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1032_DPP-NEXT:    ds_max_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    buffer_gl0_inv
 ; GFX1032_DPP-NEXT:  .LBB23_2:
 ; GFX1032_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1032_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1032_DPP-NEXT:    s_endpgm
 ;
 ; GFX1164_DPP-LABEL: max_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    s_mov_b32 s4, 0
-; GFX1164_DPP-NEXT:    s_brev_b32 s5, 1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, s4
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, s5
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x80000000, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x80000000, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 31
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s5
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s8, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s9, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s9, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1164_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB23_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1164_DPP-NEXT:    ds_max_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1164_DPP-NEXT:    ds_max_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB23_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, s[0:1], v[9:10]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1164_DPP-NEXT:    v_cmp_gt_i64_e32 vcc, s[0:1], v[7:8]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: max_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    s_mov_b32 s0, 0
-; GFX1132_DPP-NEXT:    s_brev_b32 s1, 1
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v2, s1 :: v_dual_mov_b32 v1, s0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x80000000, 0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x80000000, 0, s4
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s4
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1132_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s6, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1132_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_cbranch_execz .LBB23_2
 ; GFX1132_DPP-NEXT:  ; %bb.1:
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v10, s1 :: v_dual_mov_b32 v9, s0
-; GFX1132_DPP-NEXT:    ds_max_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v8, s1 :: v_dual_mov_b32 v7, s0
+; GFX1132_DPP-NEXT:    ds_max_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    buffer_gl0_inv
 ; GFX1132_DPP-NEXT:  .LBB23_2:
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1132_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1132_DPP-NEXT:    v_cmp_gt_i64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1132_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1132_DPP-NEXT:    s_endpgm
 entry:
   %lane = call i32 @llvm.amdgcn.workitem.id.x()
@@ -13566,65 +13469,61 @@ define amdgpu_kernel void @min_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, 0
 ; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX8_DPP-NEXT:    s_brev_b32 s7, -2
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, s6
-; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v1, -2
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, s7
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, v1, 0, s[0:1]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v2, -2
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v4, v2, 0, s[0:1]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v1, -1
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_readlane_b32 s5, v2, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s4, v1, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX8_DPP-NEXT:    v_readlane_b32 s5, v4, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s4, v3, 63
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
@@ -13638,19 +13537,20 @@ define amdgpu_kernel void @min_i64_varying(ptr addrspace(1) %out) {
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB26_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s1, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s0, v7
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX8_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[0:1], v[7:8]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s1
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v8
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v7
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v1
+; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX8_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[4:5], v[7:8]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s0
-; GFX8_DPP-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
+; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v7, v7, v0, vcc
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: min_i64_varying:
@@ -13659,65 +13559,61 @@ define amdgpu_kernel void @min_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, 0
 ; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX9_DPP-NEXT:    s_brev_b32 s7, -2
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, s6
-; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v1, -2
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, s7
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, v1, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v2, -2
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v4, v2, 0, s[0:1]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v1, -1
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_readlane_b32 s5, v2, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s4, v1, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
+; GFX9_DPP-NEXT:    v_readlane_b32 s5, v4, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s4, v3, 63
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v1, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
@@ -13730,426 +13626,423 @@ define amdgpu_kernel void @min_i64_varying(ptr addrspace(1) %out) {
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB26_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX9_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[0:1], v[7:8]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s1
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v8
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v7
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v1
+; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX9_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[4:5], v[7:8]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9_DPP-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9_DPP-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9_DPP-NEXT:    s_mov_b32 s2, -1
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v7, v7, v0, vcc
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: min_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1064_DPP-NEXT:    s_brev_b32 s7, -2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, s6
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, s7
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x7fffffff, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x7fffffff, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_bfrev_b32_e32 v5, -2
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v3, 31
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s10, v4, 47
-; GFX1064_DPP-NEXT:    v_readlane_b32 s11, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s7, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s8, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
-; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
+; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s10, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s11, 48
-; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[8:9]
-; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
-; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s9, 48
+; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
+; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
+; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB26_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1064_DPP-NEXT:    ds_min_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1064_DPP-NEXT:    ds_min_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB26_2:
 ; GFX1064_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1064_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[0:1], v[9:10]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1064_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[0:1], v[7:8]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1064_DPP-NEXT:    s_endpgm
 ;
 ; GFX1032_DPP-LABEL: min_i64_varying:
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1032_DPP-NEXT:    s_brev_b32 s7, -2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, s6
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, s7
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x7fffffff, 0, s4
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s7, v3, 15
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x7fffffff, 0, s4
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s4
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_bfrev_b32_e32 v5, -2
+; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s7, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
+; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1032_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1032_DPP-NEXT:    s_cbranch_execz .LBB26_2
 ; GFX1032_DPP-NEXT:  ; %bb.1:
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1032_DPP-NEXT:    ds_min_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1032_DPP-NEXT:    ds_min_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    buffer_gl0_inv
 ; GFX1032_DPP-NEXT:  .LBB26_2:
 ; GFX1032_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1032_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1032_DPP-NEXT:    s_endpgm
 ;
 ; GFX1164_DPP-LABEL: min_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1164_DPP-NEXT:    s_brev_b32 s7, -2
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, s6
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, s7
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x7fffffff, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x7fffffff, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 31
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s5
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_bfrev_b32_e32 v5, -2
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1164_DPP-NEXT:    v_readlane_b32 s10, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s11, v3, 47
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s7, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s8, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[8:9], -1
+; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s10, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s11, 48
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[8:9]
-; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s9, 48
+; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[6:7]
+; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
+; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1164_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB26_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1164_DPP-NEXT:    ds_min_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1164_DPP-NEXT:    ds_min_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB26_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[0:1], v[9:10]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1164_DPP-NEXT:    v_cmp_lt_i64_e32 vcc, s[0:1], v[7:8]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: min_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    s_brev_b32 s7, -2
-; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v1, s6 :: v_dual_mov_b32 v2, s7
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v4, 0x7fffffff, 0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, 0x7fffffff, 0, s4
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s4
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v6, -2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v4, -2
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s7, v3, 15
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_bfrev_b32_e32 v5, -2
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, -1 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1132_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s7, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
+; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1132_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_cbranch_execz .LBB26_2
 ; GFX1132_DPP-NEXT:  ; %bb.1:
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v10, s1 :: v_dual_mov_b32 v9, s0
-; GFX1132_DPP-NEXT:    ds_min_rtn_i64 v[9:10], v0, v[9:10]
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v8, s1 :: v_dual_mov_b32 v7, s0
+; GFX1132_DPP-NEXT:    ds_min_rtn_i64 v[7:8], v0, v[7:8]
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    buffer_gl0_inv
 ; GFX1132_DPP-NEXT:  .LBB26_2:
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1132_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1132_DPP-NEXT:    v_cmp_lt_i64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1132_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1132_DPP-NEXT:    s_endpgm
 entry:
   %lane = call i32 @llvm.amdgcn.workitem.id.x()
@@ -15395,587 +15288,581 @@ define amdgpu_kernel void @umax_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX8_DPP-LABEL: umax_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v5, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[0:1]
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX8_DPP-NEXT:    v_readlane_b32 s5, v2, 63
 ; GFX8_DPP-NEXT:    v_readlane_b32 s4, v1, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
+; GFX8_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB29_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, s4
 ; GFX8_DPP-NEXT:    s_mov_b32 m0, -1
-; GFX8_DPP-NEXT:    ds_max_rtn_u64 v[7:8], v9, v[7:8]
+; GFX8_DPP-NEXT:    ds_max_rtn_u64 v[5:6], v7, v[5:6]
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB29_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v7
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v6
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[4:5], v[7:8]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[4:5], v[5:6]
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s5
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v6, v6, v0, vcc
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v7, v7, v0, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v5, v5, v0, vcc
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: umax_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v5, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v5, exec_hi, v5
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[0:1]
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX9_DPP-NEXT:    v_readlane_b32 s5, v2, 63
 ; GFX9_DPP-NEXT:    v_readlane_b32 s4, v1, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v5
+; GFX9_DPP-NEXT:    ; implicit-def: $vgpr5_vgpr6
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB29_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX9_DPP-NEXT:    ds_max_rtn_u64 v[7:8], v9, v[7:8]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX9_DPP-NEXT:    ds_max_rtn_u64 v[5:6], v7, v[5:6]
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB29_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v6
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
 ; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[4:5], v[7:8]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[4:5], v[5:6]
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s5
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v6, v6, v0, vcc
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX9_DPP-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX9_DPP-NEXT:    s_mov_b32 s2, -1
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v7, v7, v0, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v5, v5, v0, vcc
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[0:3], 0
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[5:6], off, s[0:3], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: umax_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s8, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s9, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s9, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB29_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1064_DPP-NEXT:    ds_max_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1064_DPP-NEXT:    ds_max_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB29_2:
 ; GFX1064_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1064_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[9:10]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1064_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[7:8]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1064_DPP-NEXT:    s_endpgm
 ;
 ; GFX1032_DPP-LABEL: umax_i64_varying:
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, 0, s4
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s6, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1032_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1032_DPP-NEXT:    s_cbranch_execz .LBB29_2
 ; GFX1032_DPP-NEXT:  ; %bb.1:
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1032_DPP-NEXT:    ds_max_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1032_DPP-NEXT:    ds_max_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    buffer_gl0_inv
 ; GFX1032_DPP-NEXT:  .LBB29_2:
 ; GFX1032_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1032_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1032_DPP-NEXT:    s_endpgm
 ;
 ; GFX1164_DPP-LABEL: umax_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 31
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s5
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s8, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s9, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s9, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1164_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB29_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1164_DPP-NEXT:    ds_max_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1164_DPP-NEXT:    ds_max_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB29_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[9:10]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1164_DPP-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[7:8]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: umax_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, 0
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v4, 0, 0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, 0, 0, s4
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v3, 0
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, 0, v0, s4
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v5, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v3, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v5, 0
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_mov_b32 v3, 0
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, 0 :: v_dual_cndmask_b32 v2, v4, v2
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, 0 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1132_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s6, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1132_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_cbranch_execz .LBB29_2
 ; GFX1132_DPP-NEXT:  ; %bb.1:
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v10, s1 :: v_dual_mov_b32 v9, s0
-; GFX1132_DPP-NEXT:    ds_max_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v8, s1 :: v_dual_mov_b32 v7, s0
+; GFX1132_DPP-NEXT:    ds_max_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    buffer_gl0_inv
 ; GFX1132_DPP-NEXT:  .LBB29_2:
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1132_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1132_DPP-NEXT:    v_cmp_gt_u64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1132_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1132_DPP-NEXT:    s_endpgm
 entry:
   %lane = call i32 @llvm.amdgcn.workitem.id.x()
@@ -17221,585 +17108,579 @@ define amdgpu_kernel void @umin_i64_varying(ptr addrspace(1) %out) {
 ;
 ; GFX8_DPP-LABEL: umin_i64_varying:
 ; GFX8_DPP:       ; %bb.0: ; %entry
-; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX8_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX8_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
 ; GFX8_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, 0, s[0:1]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s[0:1]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v1, -1
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, 0, s[0:1]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX8_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX8_DPP-NEXT:    s_nop 0
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX8_DPP-NEXT:    v_readlane_b32 s5, v2, 63
-; GFX8_DPP-NEXT:    v_readlane_b32 s4, v1, 63
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v4, v4, v2, vcc
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v2, -1
+; GFX8_DPP-NEXT:    v_readlane_b32 s5, v3, 63
+; GFX8_DPP-NEXT:    v_readlane_b32 s4, v4, 63
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v2, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX8_DPP-NEXT:    v_mov_b32_dpp v1, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX8_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX8_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX8_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GFX8_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX8_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX8_DPP-NEXT:    s_cbranch_execz .LBB32_2
 ; GFX8_DPP-NEXT:  ; %bb.1:
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, s5
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, s4
 ; GFX8_DPP-NEXT:    s_mov_b32 m0, -1
-; GFX8_DPP-NEXT:    ds_min_rtn_u64 v[7:8], v9, v[7:8]
+; GFX8_DPP-NEXT:    ds_min_rtn_u64 v[6:7], v8, v[6:7]
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8_DPP-NEXT:  .LBB32_2:
 ; GFX8_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s1, v8
-; GFX8_DPP-NEXT:    v_readfirstlane_b32 s0, v7
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX8_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[0:1], v[7:8]
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s1
-; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
-; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s0
-; GFX8_DPP-NEXT:    s_mov_b32 s7, 0xf000
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s5, v7
+; GFX8_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v6, v1
+; GFX8_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v7, v2
+; GFX8_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7]
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v7, v7, v0, vcc
+; GFX8_DPP-NEXT:    v_mov_b32_e32 v0, s4
+; GFX8_DPP-NEXT:    s_mov_b32 s3, 0xf000
+; GFX8_DPP-NEXT:    s_mov_b32 s2, -1
+; GFX8_DPP-NEXT:    v_cndmask_b32_e32 v6, v6, v0, vcc
 ; GFX8_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
+; GFX8_DPP-NEXT:    buffer_store_dwordx2 v[6:7], off, s[0:3], 0
 ; GFX8_DPP-NEXT:    s_endpgm
 ;
 ; GFX9_DPP-LABEL: umin_i64_varying:
 ; GFX9_DPP:       ; %bb.0: ; %entry
-; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v7
+; GFX9_DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9_DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v6
 ; GFX9_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, 0, s[0:1]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, v0, s[0:1]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v1, -1
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
 ; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, 0, s[0:1]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
-; GFX9_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, v3
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v4
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v4, -1
 ; GFX9_DPP-NEXT:    s_nop 0
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
-; GFX9_DPP-NEXT:    v_readlane_b32 s5, v2, 63
-; GFX9_DPP-NEXT:    v_readlane_b32 s4, v1, 63
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v3, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[2:3], v[4:5]
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v4, v4, v2, vcc
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v2, -1
+; GFX9_DPP-NEXT:    v_readlane_b32 s5, v3, 63
+; GFX9_DPP-NEXT:    v_readlane_b32 s4, v4, 63
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v2, v3 wave_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9_DPP-NEXT:    v_mov_b32_dpp v1, v4 wave_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX9_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
-; GFX9_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
+; GFX9_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GFX9_DPP-NEXT:    ; implicit-def: $vgpr6_vgpr7
 ; GFX9_DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9_DPP-NEXT:    s_cbranch_execz .LBB32_2
 ; GFX9_DPP-NEXT:  ; %bb.1:
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX9_DPP-NEXT:    ds_min_rtn_u64 v[7:8], v9, v[7:8]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, s5
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, s4
+; GFX9_DPP-NEXT:    ds_min_rtn_u64 v[6:7], v8, v[6:7]
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9_DPP-NEXT:  .LBB32_2:
 ; GFX9_DPP-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s1, v8
-; GFX9_DPP-NEXT:    v_readfirstlane_b32 s0, v7
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX9_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[0:1], v[7:8]
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s1
-; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v8, v8, v0, vcc
-; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9_DPP-NEXT:    s_mov_b32 s7, 0xf000
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s5, v7
+; GFX9_DPP-NEXT:    v_readfirstlane_b32 s4, v6
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v6, v1
+; GFX9_DPP-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x24
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v7, v2
+; GFX9_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7]
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v7, v7, v0, vcc
+; GFX9_DPP-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9_DPP-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9_DPP-NEXT:    s_mov_b32 s2, -1
+; GFX9_DPP-NEXT:    v_cndmask_b32_e32 v6, v6, v0, vcc
 ; GFX9_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
+; GFX9_DPP-NEXT:    buffer_store_dwordx2 v[6:7], off, s[0:3], 0
 ; GFX9_DPP-NEXT:    s_endpgm
 ;
 ; GFX1064_DPP-LABEL: umin_i64_varying:
 ; GFX1064_DPP:       ; %bb.0: ; %entry
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v1, -1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v2, -1
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v4, -1, 0, s[0:1]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s[0:1]
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1064_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s4
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, 0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1064_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 31
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v6, s5
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1064_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1064_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1064_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1064_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1064_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1064_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1064_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1064_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1064_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1064_DPP-NEXT:    v_writelane_b32 v2, s8, 48
-; GFX1064_DPP-NEXT:    v_writelane_b32 v1, s9, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1064_DPP-NEXT:    v_writelane_b32 v4, s9, 48
 ; GFX1064_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1064_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1064_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1064_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1064_DPP-NEXT:    s_cbranch_execz .LBB32_2
 ; GFX1064_DPP-NEXT:  ; %bb.1:
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1064_DPP-NEXT:    ds_min_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1064_DPP-NEXT:    ds_min_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064_DPP-NEXT:    buffer_gl0_inv
 ; GFX1064_DPP-NEXT:  .LBB32_2:
 ; GFX1064_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1064_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1064_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1064_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1064_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1064_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1064_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[0:1], v[9:10]
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1064_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[0:1], v[7:8]
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1064_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1064_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1064_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1064_DPP-NEXT:    s_endpgm
 ;
 ; GFX1032_DPP-LABEL: umin_i64_varying:
 ; GFX1032_DPP:       ; %bb.0: ; %entry
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v1, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v2, -1
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v4, -1, 0, s4
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s4
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1032_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc_lo
-; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1032_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, 0, s4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1032_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc_lo
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1032_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1032_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1032_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1032_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1032_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
-; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1032_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1032_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1032_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1032_DPP-NEXT:    v_writelane_b32 v1, s6, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1032_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1032_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1032_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1032_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1032_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1032_DPP-NEXT:    s_cbranch_execz .LBB32_2
 ; GFX1032_DPP-NEXT:  ; %bb.1:
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1032_DPP-NEXT:    ds_min_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1032_DPP-NEXT:    ds_min_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032_DPP-NEXT:    buffer_gl0_inv
 ; GFX1032_DPP-NEXT:  .LBB32_2:
 ; GFX1032_DPP-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX1032_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1032_DPP-NEXT:    s_load_dwordx2 s[4:5], s[2:3], 0x24
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1032_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1032_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1032_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1032_DPP-NEXT:    s_mov_b32 s7, 0x31016000
-; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1032_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1032_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1032_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[9:10], off, s[4:7], 0
+; GFX1032_DPP-NEXT:    buffer_store_dwordx2 v[7:8], off, s[4:7], 0
 ; GFX1032_DPP-NEXT:    s_endpgm
 ;
 ; GFX1164_DPP-LABEL: umin_i64_varying:
 ; GFX1164_DPP:       ; %bb.0: ; %entry
-; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v1, -1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v2, -1
-; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v4, -1, 0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s[0:1]
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, 0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s[0:1]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v2
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[7:8]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[5:6]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1164_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
+; GFX1164_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v3, -1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 31
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_4)
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 31
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 31
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, s4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v6, s5
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s5
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[3:4], v[5:6]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v5, -1
+; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(SALU_CYCLE_1)
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, v[1:2], v[3:4]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v2, v4, v2, vcc
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v4, -1
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164_DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v4, 15
-; GFX1164_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v3, 15
-; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v4, 31
-; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v3, 31
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s4, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v3, 63
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s5, 16
-; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v4, 63
-; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v4, 47
-; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v3, 47
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s6, 32
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s7, 32
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v2, 15
+; GFX1164_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v1, 15
+; GFX1164_DPP-NEXT:    v_readlane_b32 s6, v2, 31
+; GFX1164_DPP-NEXT:    v_readlane_b32 s7, v1, 31
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s4, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s4, v1, 63
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s5, 16
+; GFX1164_DPP-NEXT:    v_readlane_b32 s5, v2, 63
+; GFX1164_DPP-NEXT:    v_readlane_b32 s8, v2, 47
+; GFX1164_DPP-NEXT:    v_readlane_b32 s9, v1, 47
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s6, 32
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s7, 32
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v9, exec_hi, v0
+; GFX1164_DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1164_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164_DPP-NEXT:    s_or_saveexec_b64 s[6:7], -1
 ; GFX1164_DPP-NEXT:    s_mov_b64 s[0:1], s[4:5]
-; GFX1164_DPP-NEXT:    v_writelane_b32 v2, s8, 48
-; GFX1164_DPP-NEXT:    v_writelane_b32 v1, s9, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v5, s8, 48
+; GFX1164_DPP-NEXT:    v_writelane_b32 v4, s9, 48
 ; GFX1164_DPP-NEXT:    s_mov_b64 exec, s[6:7]
-; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
+; GFX1164_DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1164_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1164_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1164_DPP-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GFX1164_DPP-NEXT:    s_cbranch_execz .LBB32_2
 ; GFX1164_DPP-NEXT:  ; %bb.1:
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, s1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, s0
-; GFX1164_DPP-NEXT:    ds_min_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, s1
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, s0
+; GFX1164_DPP-NEXT:    ds_min_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164_DPP-NEXT:    buffer_gl0_inv
 ; GFX1164_DPP-NEXT:  .LBB32_2:
 ; GFX1164_DPP-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; GFX1164_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1164_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1164_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1164_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1164_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1164_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[0:1], v[9:10]
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc
-; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc
+; GFX1164_DPP-NEXT:    v_cmp_lt_u64_e32 vcc, s[0:1], v[7:8]
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc
+; GFX1164_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc
 ; GFX1164_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1164_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1164_DPP-NEXT:    s_endpgm
 ;
 ; GFX1132_DPP-LABEL: umin_i64_varying:
 ; GFX1132_DPP:       ; %bb.0: ; %entry
-; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v1, -1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v2, -1
-; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1132_DPP-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v4, -1, 0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v3, -1, v0, s4
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v2, -1, 0, s4
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, -1 :: v_dual_mov_b32 v3, -1
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v1, -1, v0, s4
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, -1 :: v_dual_mov_b32 v5, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, -1 :: v_dual_mov_b32 v3, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v7, v1 :: v_dual_mov_b32 v8, v2
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v6, -1 :: v_dual_mov_b32 v5, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v3 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v2 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v1 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, v1 :: v_dual_mov_b32 v6, v2
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v4, v2 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, -1 :: v_dual_mov_b32 v3, -1
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v2 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[7:8]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v8, v4 :: v_dual_cndmask_b32 v3, v7, v3
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v1 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[5:6]
+; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v2, v6, v2 :: v_dual_cndmask_b32 v1, v5, v1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v7, v4, -1, -1
-; GFX1132_DPP-NEXT:    v_permlanex16_b32 v8, v3, -1, -1
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v5, v2, -1, -1
+; GFX1132_DPP-NEXT:    v_permlanex16_b32 v6, v1, -1, -1
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v6, v7 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v8 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[3:4], v[5:6]
-; GFX1132_DPP-NEXT:    v_dual_cndmask_b32 v4, v6, v4 :: v_dual_cndmask_b32 v3, v5, v3
-; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v4, 31
-; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v3, 15
-; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v3, 31
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v2, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v4, 15
-; GFX1132_DPP-NEXT:    v_mov_b32_dpp v1, v3 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v5 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v3, v6 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, v[1:2], v[3:4]
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v5, -1 :: v_dual_cndmask_b32 v2, v4, v2
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v4, -1 :: v_dual_cndmask_b32 v1, v3, v1
+; GFX1132_DPP-NEXT:    v_readlane_b32 s1, v2, 31
+; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v5, v2 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX1132_DPP-NEXT:    v_readlane_b32 s6, v1, 15
+; GFX1132_DPP-NEXT:    v_readlane_b32 s0, v1, 31
+; GFX1132_DPP-NEXT:    v_readlane_b32 s5, v2, 15
+; GFX1132_DPP-NEXT:    v_mov_b32_dpp v4, v1 row_shr:1 row_mask:0xf bank_mask:0xf
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v9, exec_lo, 0
+; GFX1132_DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
 ; GFX1132_DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132_DPP-NEXT:    s_or_saveexec_b32 s4, -1
-; GFX1132_DPP-NEXT:    v_writelane_b32 v2, s5, 16
-; GFX1132_DPP-NEXT:    v_writelane_b32 v1, s6, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v5, s5, 16
+; GFX1132_DPP-NEXT:    v_writelane_b32 v4, s6, 16
 ; GFX1132_DPP-NEXT:    s_mov_b32 exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_mov_b32 s6, -1
-; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v9
-; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr9_vgpr10
+; GFX1132_DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
+; GFX1132_DPP-NEXT:    ; implicit-def: $vgpr7_vgpr8
 ; GFX1132_DPP-NEXT:    s_and_saveexec_b32 s4, vcc_lo
 ; GFX1132_DPP-NEXT:    s_cbranch_execz .LBB32_2
 ; GFX1132_DPP-NEXT:  ; %bb.1:
-; GFX1132_DPP-NEXT:    v_dual_mov_b32 v10, s1 :: v_dual_mov_b32 v9, s0
-; GFX1132_DPP-NEXT:    ds_min_rtn_u64 v[9:10], v0, v[9:10]
+; GFX1132_DPP-NEXT:    v_dual_mov_b32 v8, s1 :: v_dual_mov_b32 v7, s0
+; GFX1132_DPP-NEXT:    ds_min_rtn_u64 v[7:8], v0, v[7:8]
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132_DPP-NEXT:    buffer_gl0_inv
 ; GFX1132_DPP-NEXT:  .LBB32_2:
 ; GFX1132_DPP-NEXT:    s_or_b32 exec_lo, exec_lo, s4
 ; GFX1132_DPP-NEXT:    s_load_b64 s[4:5], s[2:3], 0x24
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v10
-; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v9
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v9, v1
-; GFX1132_DPP-NEXT:    v_mov_b32_e32 v10, v2
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s1, v8
+; GFX1132_DPP-NEXT:    v_readfirstlane_b32 s0, v7
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v7, v4
+; GFX1132_DPP-NEXT:    v_mov_b32_e32 v8, v5
 ; GFX1132_DPP-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1132_DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, s[0:1], v[9:10]
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v10, v10, s1, vcc_lo
-; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v9, v9, s0, vcc_lo
+; GFX1132_DPP-NEXT:    v_cmp_lt_u64_e32 vcc_lo, s[0:1], v[7:8]
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v8, v8, s1, vcc_lo
+; GFX1132_DPP-NEXT:    v_cndmask_b32_e64 v7, v7, s0, vcc_lo
 ; GFX1132_DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132_DPP-NEXT:    buffer_store_b64 v[9:10], off, s[4:7], 0
+; GFX1132_DPP-NEXT:    buffer_store_b64 v[7:8], off, s[4:7], 0
 ; GFX1132_DPP-NEXT:    s_endpgm
 entry:
   %lane = call i32 @llvm.amdgcn.workitem.id.x()
diff --git a/llvm/test/CodeGen/AMDGPU/dpp64_combine.ll b/llvm/test/CodeGen/AMDGPU/dpp64_combine.ll
index 60ce730c3eed33..c7422786d344ee 100644
--- a/llvm/test/CodeGen/AMDGPU/dpp64_combine.ll
+++ b/llvm/test/CodeGen/AMDGPU/dpp64_combine.ll
@@ -1,7 +1,7 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,DPP64,GFX90A
 ; RUN: llc -mtriple=amdgcn -mcpu=gfx940 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,DPP64,DPPMOV64,GFX940
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,DPP32,GFX10PLUS,GFX10
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,DPP32,GFX10PLUS,GFX11
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,DPP32,GFX10PLUS
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,DPP32,GFX10PLUS
 
 ; GCN-LABEL: {{^}}dpp64_ceil:
 ; GCN:           global_load_{{dwordx2|b64}} [[V:v\[[0-9:]+\]]],
@@ -75,16 +75,14 @@ define amdgpu_kernel void @dpp64_div(ptr addrspace(1) %arg, i64 %in1) {
 ; GFX90A: v_add_co_u32_e32
 ; GFX90A: v_addc_co_u32_e32
 ; GFX940: v_lshl_add_u64
-; GFX10: v_mov_b32_dpp
-; GFX10: v_add_co_u32
-; GFX10: v_add_co_ci_u32_e32
-; GFX11: v_add_co_u32_e64_dpp
-; GFX11: v_add_co_ci_u32_e32
-define amdgpu_cs void @dpp64_loop(i64 %arg) {
+; GFX10PLUS: v_mov_b32_dpp
+; GFX10PLUS: v_add_co_u32
+; GFX10PLUS: v_add_co_ci_u32_e32
+define amdgpu_cs void @dpp64_loop(i64 %arg, i64 %val) {
 bb:
   br label %bb1
 bb1:
-  %i = call i64 @llvm.amdgcn.update.dpp.i64(i64 0, i64 0, i32 0, i32 0, i32 0, i1 false)
+  %i = call i64 @llvm.amdgcn.update.dpp.i64(i64 poison, i64 %val, i32 0, i32 0, i32 0, i1 false)
   %i2 = add i64 %i, %arg
   %i3 = atomicrmw add ptr addrspace(1) null, i64 %i2 monotonic, align 8
   br label %bb1
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll
index fa0ab4537fcd76..7cc78b47a72214 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fadd.ll
@@ -7307,58 +7307,56 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -7439,28 +7437,28 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
@@ -7473,13 +7471,13 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -7561,42 +7559,42 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
 ; GFX1032-DPP-NEXT:  .LBB10_2: ; %atomicrmw.start
@@ -7667,35 +7665,34 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -7709,16 +7706,16 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
@@ -7789,49 +7786,50 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v41, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
 ; GFX1132-DPP-NEXT:    s_set_inst_prefetch_distance 0x1
@@ -8885,59 +8883,57 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -8985,28 +8981,28 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
@@ -9017,27 +9013,27 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], s[2:3], s[4:5]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1064-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1064-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], v[0:1]
-; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1064-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], v[0:1]
+; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[9:10], v[11:12]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[9:10]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1064-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1064-DPP-NEXT:    s_andn2_b64 exec, exec, s[2:3]
 ; GFX1064-DPP-NEXT:    s_cbranch_execnz .LBB12_2
@@ -9073,53 +9069,53 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1032-DPP-NEXT:    s_mov_b32 s2, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1032-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1032-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], v[0:1]
-; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1032-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], v[0:1]
+; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[9:10], v[11:12]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[9:10]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1032-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1032-DPP-NEXT:    s_andn2_b32 exec_lo, exec_lo, s2
 ; GFX1032-DPP-NEXT:    s_cbranch_execnz .LBB12_2
@@ -9145,35 +9141,34 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -9184,32 +9179,31 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    v_permlane64_b32 v4, v2
 ; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -9236,59 +9230,60 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_one_a
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -10309,59 +10304,57 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -10409,28 +10402,28 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
@@ -10441,27 +10434,27 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], s[2:3], s[4:5]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1064-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1064-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], v[0:1]
-; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1064-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], v[0:1]
+; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[9:10], v[11:12]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[9:10]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1064-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1064-DPP-NEXT:    s_andn2_b64 exec, exec, s[2:3]
 ; GFX1064-DPP-NEXT:    s_cbranch_execnz .LBB14_2
@@ -10497,53 +10490,53 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1032-DPP-NEXT:    s_mov_b32 s2, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1032-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1032-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], v[0:1]
-; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1032-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], v[0:1]
+; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[9:10], v[11:12]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[9:10]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1032-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1032-DPP-NEXT:    s_andn2_b32 exec_lo, exec_lo, s2
 ; GFX1032-DPP-NEXT:    s_cbranch_execnz .LBB14_2
@@ -10569,35 +10562,34 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -10608,32 +10600,31 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    v_permlane64_b32 v4, v2
 ; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -10660,59 +10651,60 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -11215,59 +11207,57 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -11315,28 +11305,28 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
@@ -11347,27 +11337,27 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], s[2:3], s[4:5]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1064-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1064-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], v[0:1]
-; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1064-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], v[0:1]
+; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[9:10], v[11:12]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[9:10]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1064-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1064-DPP-NEXT:    s_andn2_b64 exec, exec, s[2:3]
 ; GFX1064-DPP-NEXT:    s_cbranch_execnz .LBB15_2
@@ -11403,53 +11393,53 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1032-DPP-NEXT:    s_mov_b32 s2, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1032-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1032-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], v[0:1]
-; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1032-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], v[0:1]
+; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[9:10], v[11:12]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[9:10]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1032-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1032-DPP-NEXT:    s_andn2_b32 exec_lo, exec_lo, s2
 ; GFX1032-DPP-NEXT:    s_cbranch_execnz .LBB15_2
@@ -11475,35 +11465,34 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -11514,32 +11503,31 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    v_permlane64_b32 v4, v2
 ; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -11566,59 +11554,60 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -13286,58 +13275,56 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -13418,28 +13405,28 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
@@ -13452,13 +13439,13 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -13540,42 +13527,42 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
 ; GFX1032-DPP-NEXT:  .LBB17_2: ; %atomicrmw.start
@@ -13646,35 +13633,34 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -13688,16 +13674,16 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
@@ -13768,49 +13754,50 @@ define amdgpu_kernel void @global_atomic_fadd_double_uni_address_div_value_defau
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v41, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
 ; GFX1132-DPP-NEXT:    s_set_inst_prefetch_distance 0x1
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll
index 9b9dd744945b15..8e7181a0cf4495 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmax.ll
@@ -4709,65 +4709,63 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -4850,33 +4848,33 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
@@ -4892,13 +4890,13 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -4982,48 +4980,48 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -5096,42 +5094,41 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -5148,16 +5145,16 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
-; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
@@ -5230,57 +5227,55 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, 0x7ff80000 :: v_dual_mov_b32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v3, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
@@ -6155,66 +6150,64 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0x7ff80000
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v4, 0x7ff80000
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
+; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB9_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -6264,33 +6257,33 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v4, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x7ff80000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
-; GFX1064-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
 ; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
@@ -6304,15 +6297,15 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[5:6], v[3:4]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB9_2
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_atomic_fmax_x2 v2, v[0:1], s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB9_2:
@@ -6347,47 +6340,47 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v4, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x7ff80000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
-; GFX1032-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
 ; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
 ; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB9_2
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_atomic_fmax_x2 v2, v[0:1], s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB9_2:
@@ -6412,42 +6405,41 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v3, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x7ff80000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v5, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v5, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v5, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
-; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -6461,35 +6453,34 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
 ; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
-; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB9_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1]
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB9_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[8:9], v[8:9]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -6516,70 +6507,68 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_one_a
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v3, 0x7ff80000
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v3, 0x7ff80000 :: v_dual_mov_b32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x7ff80000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
-; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v5, 0x7ff80000 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v5, 0x7ff80000 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v5, 0x7ff80000 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB9_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1]
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB9_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[8:9], v[8:9]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -8205,65 +8194,63 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -8346,33 +8333,33 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
@@ -8388,13 +8375,13 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -8478,48 +8465,48 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -8592,42 +8579,41 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -8644,16 +8630,16 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
-; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
@@ -8726,57 +8712,55 @@ define amdgpu_kernel void @global_atomic_fmax_double_uni_address_div_value_defau
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, 0x7ff80000 :: v_dual_mov_b32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v3, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll
index 21a65851db1d55..b95b52168625dd 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fmin.ll
@@ -4709,65 +4709,63 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -4850,33 +4848,33 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
@@ -4892,13 +4890,13 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -4982,48 +4980,48 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -5096,42 +5094,41 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -5148,16 +5145,16 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
-; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
@@ -5230,57 +5227,55 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, 0x7ff80000 :: v_dual_mov_b32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v3, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB7_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
@@ -6155,66 +6150,64 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0x7ff80000
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v4, 0x7ff80000
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
-; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
+; GFX9-DPP-NEXT:    v_min_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB9_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -6264,33 +6257,33 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v4, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x7ff80000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
-; GFX1064-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1064-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1064-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1064-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
 ; GFX1064-DPP-NEXT:    v_min_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1064-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1064-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1064-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
@@ -6304,15 +6297,15 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    v_min_f64 v[3:4], v[5:6], v[3:4]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB9_2
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_atomic_fmin_x2 v2, v[0:1], s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB9_2:
@@ -6347,47 +6340,47 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v4, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x7ff80000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
-; GFX1032-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1032-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1032-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[7:8], v[7:8], v[7:8]
-; GFX1032-DPP-NEXT:    v_min_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[3:4], v[3:4], v[3:4]
 ; GFX1032-DPP-NEXT:    v_min_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1032-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1032-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v6, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
+; GFX1032-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[5:6], v[5:6], v[5:6]
 ; GFX1032-DPP-NEXT:    v_min_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB9_2
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_atomic_fmin_x2 v2, v[0:1], s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB9_2:
@@ -6412,42 +6405,41 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v3, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x7ff80000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1164-DPP-NEXT:    v_min_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
+; GFX1164-DPP-NEXT:    v_min_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v5, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1164-DPP-NEXT:    v_min_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1164-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v5, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1164-DPP-NEXT:    v_min_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1164-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v5, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
-; GFX1164-DPP-NEXT:    v_min_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1164-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -6461,35 +6453,34 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
 ; GFX1164-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
-; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB9_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1]
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB9_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_max_f64 v[6:7], v[8:9], v[8:9]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_min_f64 v[6:7], v[6:7], v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -6516,70 +6507,68 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_one_a
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v3, 0x7ff80000
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v3, 0x7ff80000 :: v_dual_mov_b32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x7ff80000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
-; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[6:7], v[6:7]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[2:3], v[2:3], v[2:3]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_min_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v5, 0x7ff80000 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1132-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v5, 0x7ff80000 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1132-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v5, 0x7ff80000 :: v_dual_mov_b32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
+; GFX1132-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[4:5], v[4:5], v[4:5]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_min_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB9_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    v_max_f64 v[0:1], v[0:1], v[0:1]
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB9_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_max_f64 v[6:7], v[8:9], v[8:9]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_min_f64 v[6:7], v[6:7], v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -8205,65 +8194,63 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0x7ff80000
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, 0x7ff80000
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
 ; GFX9-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX9-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -8346,33 +8333,33 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1064-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1064-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
@@ -8388,13 +8375,13 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -8478,48 +8465,48 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1032-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
 ; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1032-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -8592,42 +8579,41 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1164-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, 0x7ff80000
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1164-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -8644,16 +8630,16 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
-; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
@@ -8726,57 +8712,55 @@ define amdgpu_kernel void @global_atomic_fmin_double_uni_address_div_value_defau
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v9, 0x7ff80000
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, 0x7ff80000 :: v_dual_mov_b32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x7ff80000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_max_f64 v[12:13], v[12:13], v[12:13]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_min_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    v_max_f64 v[8:9], v[8:9], v[8:9]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, 0x7ff80000 :: v_dual_mov_b32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
+; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_max_f64 v[10:11], v[10:11], v[10:11]
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_min_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v3, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v3, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB11_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    v_max_f64 v[41:42], v[3:4], v[3:4]
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
diff --git a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll
index fcd5d0dc497e67..ae4ca7b7356ef9 100644
--- a/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll
+++ b/llvm/test/CodeGen/AMDGPU/global_atomics_scan_fsub.ll
@@ -7635,58 +7635,56 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -7767,28 +7765,28 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
@@ -7801,13 +7799,13 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -7889,42 +7887,42 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
 ; GFX1032-DPP-NEXT:  .LBB10_2: ; %atomicrmw.start
@@ -7995,35 +7993,34 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -8037,16 +8034,16 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
@@ -8117,49 +8114,50 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v41, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB10_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
 ; GFX1132-DPP-NEXT:    s_set_inst_prefetch_distance 0x1
@@ -9212,59 +9210,57 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -9312,28 +9308,28 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
@@ -9344,27 +9340,27 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], s[2:3], s[4:5]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1064-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1064-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], -v[0:1]
-; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1064-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], -v[0:1]
+; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[9:10], v[11:12]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[9:10]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1064-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1064-DPP-NEXT:    s_andn2_b64 exec, exec, s[2:3]
 ; GFX1064-DPP-NEXT:    s_cbranch_execnz .LBB12_2
@@ -9400,53 +9396,53 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1032-DPP-NEXT:    s_mov_b32 s2, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1032-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1032-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], -v[0:1]
-; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1032-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], -v[0:1]
+; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[9:10], v[11:12]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[9:10]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1032-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1032-DPP-NEXT:    s_andn2_b32 exec_lo, exec_lo, s2
 ; GFX1032-DPP-NEXT:    s_cbranch_execnz .LBB12_2
@@ -9472,35 +9468,34 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -9511,32 +9506,31 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX1164-DPP-NEXT:    v_permlane64_b32 v4, v2
 ; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], -v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], -v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -9563,59 +9557,60 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_one_a
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB12_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB12_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], -v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], -v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -10636,59 +10631,57 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -10736,28 +10729,28 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
@@ -10768,27 +10761,27 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], s[2:3], s[4:5]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1064-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1064-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], -v[0:1]
-; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1064-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], -v[0:1]
+; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[9:10], v[11:12]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[9:10]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1064-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1064-DPP-NEXT:    s_andn2_b64 exec, exec, s[2:3]
 ; GFX1064-DPP-NEXT:    s_cbranch_execnz .LBB14_2
@@ -10824,53 +10817,53 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1032-DPP-NEXT:    s_mov_b32 s2, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1032-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1032-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], -v[0:1]
-; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1032-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], -v[0:1]
+; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[9:10], v[11:12]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[9:10]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1032-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1032-DPP-NEXT:    s_andn2_b32 exec_lo, exec_lo, s2
 ; GFX1032-DPP-NEXT:    s_cbranch_execnz .LBB14_2
@@ -10896,35 +10889,34 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -10935,32 +10927,31 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    v_permlane64_b32 v4, v2
 ; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], -v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], -v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -10987,59 +10978,60 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB14_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB14_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], -v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], -v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -11542,59 +11534,57 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v4, v3, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v3, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v6, v4, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v3, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, v5
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, v6
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v3 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v4 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[7:8]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s3, v4, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s2, v3, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX9-DPP-NEXT:    s_mov_b64 s[0:1], s[2:3]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[2:3], s[34:35], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[11:12], v0, s[2:3]
@@ -11642,28 +11632,28 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
@@ -11674,27 +11664,27 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1064-DPP-NEXT:    v_add_f64 v[3:4], s[2:3], s[4:5]
 ; GFX1064-DPP-NEXT:    s_mov_b64 exec, s[0:1]
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v7, exec_hi, v0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v1, v4
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v7
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1064-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1064-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1064-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1064-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], -v[0:1]
-; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1064-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], -v[0:1]
+; GFX1064-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1064-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[9:10], v[11:12]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1064-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[7:8], v[9:10]
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1064-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1064-DPP-NEXT:    s_andn2_b64 exec, exec, s[2:3]
 ; GFX1064-DPP-NEXT:    s_cbranch_execnz .LBB15_2
@@ -11730,53 +11720,53 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v4, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v6, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v5, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, v4
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v6 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[5:6], v[5:6], v[7:8]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v4, v6 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[5:6], v[3:4]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v6, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v5, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v6, v4, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[3:4], v[3:4], v[5:6]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, v3
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v7, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1032-DPP-NEXT:    s_mov_b32 s2, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v2
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v7
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[0:1], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1032-DPP-NEXT:    global_load_dwordx2 v[11:12], v2, s[0:1]
+; GFX1032-DPP-NEXT:    global_load_dwordx2 v[9:10], v2, s[0:1]
 ; GFX1032-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1032-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_add_f64 v[9:10], v[11:12], -v[0:1]
-; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[9:10], v2, v[9:12], s[0:1] glc
+; GFX1032-DPP-NEXT:    v_add_f64 v[7:8], v[9:10], -v[0:1]
+; GFX1032-DPP-NEXT:    global_atomic_cmpswap_x2 v[7:8], v2, v[7:10], s[0:1] glc
 ; GFX1032-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[9:10], v[11:12]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v11, v9
+; GFX1032-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[7:8], v[9:10]
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v9, v7
 ; GFX1032-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1032-DPP-NEXT:    s_andn2_b32 exec_lo, exec_lo, s2
 ; GFX1032-DPP-NEXT:    s_cbranch_execnz .LBB15_2
@@ -11802,35 +11792,34 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v6, v2
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v7, v3
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
@@ -11841,32 +11830,31 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1164-DPP-NEXT:    v_permlane64_b32 v4, v2
 ; GFX1164-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1164-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v8, exec_hi, v0
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v6, exec_hi, v0
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v1, v3
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1164-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1164-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1164-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1164-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], -v[0:1]
-; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1164-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], -v[0:1]
+; GFX1164-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1164-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[8:9], v[10:11]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v11, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, v8
+; GFX1164-DPP-NEXT:    v_cmp_eq_u64_e32 vcc, v[6:7], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v9, v7
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, v6
 ; GFX1164-DPP-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1164-DPP-NEXT:    s_and_not1_b64 exec, exec, s[2:3]
@@ -11893,59 +11881,60 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_agent
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[2:3]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v3, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v5, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v4, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v6, v2 :: v_dual_mov_b32 v7, v3
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v6, v4 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v5, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v4, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v5, v3 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v7, v5 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[4:5], v[4:5], v[6:7]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v4, v2 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v2, v4 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v3, v5 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[4:5], v[2:3]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v5, v3, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v4, v2, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[2:3], v[2:3], v[4:5]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, v2
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v8, exec_lo, 0
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v1, v3
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v6, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v10, 0 :: v_dual_mov_b32 v1, v3
 ; GFX1132-DPP-NEXT:    s_mov_b32 s2, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v8
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v6
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB15_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[0:1], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1132-DPP-NEXT:    global_load_b64 v[10:11], v12, s[0:1]
+; GFX1132-DPP-NEXT:    global_load_b64 v[8:9], v10, s[0:1]
 ; GFX1132-DPP-NEXT:  .LBB15_2: ; %atomicrmw.start
 ; GFX1132-DPP-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], -v[0:1]
-; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[8:9], v12, v[8:11], s[0:1] glc
+; GFX1132-DPP-NEXT:    v_add_f64 v[6:7], v[8:9], -v[0:1]
+; GFX1132-DPP-NEXT:    global_atomic_cmpswap_b64 v[6:7], v10, v[6:9], s[0:1] glc
 ; GFX1132-DPP-NEXT:    s_waitcnt vmcnt(0)
-; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[8:9], v[10:11]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v11, v9 :: v_dual_mov_b32 v10, v8
+; GFX1132-DPP-NEXT:    v_cmp_eq_u64_e32 vcc_lo, v[6:7], v[8:9]
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v9, v7 :: v_dual_mov_b32 v8, v6
 ; GFX1132-DPP-NEXT:    s_or_b32 s2, vcc_lo, s2
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX1132-DPP-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s2
@@ -13612,58 +13601,56 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v10, 0
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v8, 1
-; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v9, v8, v1, s[0:1]
-; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v8, 0, v0, s[0:1]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
-; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:1 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v11, v9, v1, s[0:1]
+; GFX9-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v8, 0
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:1 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:2 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:2 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:4 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:4 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_shr:8 row_mask:0xf bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, v10
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v13, v11
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_shr:8 row_mask:0xf bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX9-DPP-NEXT:    v_bfrev_b32_e32 v13, 1
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v12, 0
 ; GFX9-DPP-NEXT:    s_nop 0
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v8 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v9 row_bcast:15 row_mask:0xa bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[12:13]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_bcast:15 row_mask:0xa bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
 ; GFX9-DPP-NEXT:    s_nop 1
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_bcast:31 row_mask:0xc bank_mask:0xf
-; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_bcast:31 row_mask:0xc bank_mask:0xf
+; GFX9-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
-; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
+; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX9-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v1
 ; GFX9-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
 ; GFX9-DPP-NEXT:    v_readlane_b32 s43, v9, 63
 ; GFX9-DPP-NEXT:    v_readlane_b32 s42, v8, 63
 ; GFX9-DPP-NEXT:    s_mov_b64 exec, s[0:1]
-; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX9-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX9-DPP-NEXT:  ; %bb.1:
 ; GFX9-DPP-NEXT:    s_load_dwordx2 s[44:45], s[36:37], 0x24
-; GFX9-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-DPP-NEXT:    s_mov_b64 s[46:47], 0
 ; GFX9-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[44:45]
@@ -13744,28 +13731,28 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1064-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1064-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1064-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1064-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1064-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
@@ -13778,13 +13765,13 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX1064-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v41, v8
 ; GFX1064-DPP-NEXT:    v_mov_b32_e32 v42, v9
-; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX1064-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1064-DPP-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX1064-DPP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
 ; GFX1064-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1064-DPP-NEXT:  ; %bb.1:
 ; GFX1064-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1064-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1064-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1064-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
@@ -13866,42 +13853,42 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    s_swappc_b64 s[30:31], s[6:7]
 ; GFX1032-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1032-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1032-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1032-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
 ; GFX1032-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1032-DPP-NEXT:    s_mov_b32 exec_lo, s0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1032-DPP-NEXT:    s_mov_b32 s44, 0
-; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
+; GFX1032-DPP-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX1032-DPP-NEXT:    s_and_saveexec_b32 s0, vcc_lo
 ; GFX1032-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1032-DPP-NEXT:  ; %bb.1:
 ; GFX1032-DPP-NEXT:    s_load_dwordx2 s[42:43], s[34:35], 0x24
-; GFX1032-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1032-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1032-DPP-NEXT:    global_load_dwordx2 v[1:2], v0, s[42:43]
 ; GFX1032-DPP-NEXT:  .LBB17_2: ; %atomicrmw.start
@@ -13972,35 +13959,34 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1164-DPP-NEXT:    s_or_saveexec_b64 s[0:1], -1
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s[0:1]
 ; GFX1164-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s[0:1]
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v12, v8
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v13, v9
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1164-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1164-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1164-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1164-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
 ; GFX1164-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
@@ -14014,16 +14000,16 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX1164-DPP-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1164-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_4) | instid1(VALU_DEP_2)
 ; GFX1164-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[0:1], exec
 ; GFX1164-DPP-NEXT:    s_waitcnt_depctr 0xfffe
-; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v0, exec_hi, v0
-; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1164-DPP-NEXT:    v_mbcnt_hi_u32_b32 v1, exec_hi, v0
+; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
+; GFX1164-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1164-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1164-DPP-NEXT:  ; %bb.1:
 ; GFX1164-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1164-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1164-DPP-NEXT:    s_mov_b64 s[44:45], 0
 ; GFX1164-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1164-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
@@ -14094,49 +14080,50 @@ define amdgpu_kernel void @global_atomic_fsub_double_uni_address_div_value_defau
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    s_swappc_b64 s[30:31], s[0:1]
 ; GFX1132-DPP-NEXT:    s_or_saveexec_b32 s0, -1
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v9, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v11, 0x80000000, v1, s0
 ; GFX1132-DPP-NEXT:    v_cndmask_b32_e64 v10, 0, v0, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:1 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:2 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:1 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:2 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
-; GFX1132-DPP-NEXT:    v_dual_mov_b32 v12, v8 :: v_dual_mov_b32 v13, v9
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v12, v10 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:2 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:4 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
+; GFX1132-DPP-NEXT:    v_bfrev_b32_e32 v11, 1
+; GFX1132-DPP-NEXT:    v_mov_b32_e32 v10, 0
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v11, v9 row_xmask:8 row_mask:0xf bank_mask:0xf
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v13, v11 row_xmask:4 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_add_f64 v[10:11], v[10:11], v[12:13]
+; GFX1132-DPP-NEXT:    v_mov_b32_dpp v10, v8 row_xmask:8 row_mask:0xf bank_mask:0xf
+; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v8, v10 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    v_mov_b32_dpp v9, v11 row_xmask:8 row_mask:0xf bank_mask:0xf
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[10:11], v[8:9]
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v11, v9, 0, 0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_permlanex16_b32 v10, v8, 0, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
 ; GFX1132-DPP-NEXT:    v_add_f64 v[8:9], v[8:9], v[10:11]
 ; GFX1132-DPP-NEXT:    s_mov_b32 exec_lo, s0
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v41, v8
-; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v0, exec_lo, 0
+; GFX1132-DPP-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v41, v8
+; GFX1132-DPP-NEXT:    v_mbcnt_lo_u32_b32 v1, exec_lo, 0
+; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2)
 ; GFX1132-DPP-NEXT:    v_mov_b32_e32 v42, v9
 ; GFX1132-DPP-NEXT:    s_mov_b32 s44, 0
 ; GFX1132-DPP-NEXT:    s_mov_b32 s0, exec_lo
-; GFX1132-DPP-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v0
+; GFX1132-DPP-NEXT:    v_cmpx_eq_u32_e32 0, v1
 ; GFX1132-DPP-NEXT:    s_cbranch_execz .LBB17_3
 ; GFX1132-DPP-NEXT:  ; %bb.1:
 ; GFX1132-DPP-NEXT:    s_load_b64 s[42:43], s[34:35], 0x24
-; GFX1132-DPP-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1132-DPP-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1132-DPP-NEXT:    global_load_b64 v[1:2], v0, s[42:43]
 ; GFX1132-DPP-NEXT:    s_set_inst_prefetch_distance 0x1
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll
new file mode 100644
index 00000000000000..841ec6efe8975d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll
@@ -0,0 +1,258 @@
+; RUN: llc -global-isel=0 -march=amdgcn -mcpu=gfx90a < %s | FileCheck --check-prefixes=GCN,GFX90A %s
+; RUN: llc -global-isel=1 -march=amdgcn -mcpu=gfx90a < %s | FileCheck --check-prefixes=GCN,GFX90A %s
+; RUN: llc -global-isel=0 -march=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GCN,GFX942 %s
+; RUN: llc -global-isel=1 -march=amdgcn -mcpu=gfx942 < %s | FileCheck --check-prefixes=GCN,GFX942 %s
+
+; DPP control value 337 is valid for 64-bit DPP on gfx942
+
+; GCN-LABEL: update_dpp_i64:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v2, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v3, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942:        v_mov_b64_dpp v[2:3], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx2 v[4:5], v[2:3], off
+
+define amdgpu_ps void @update_dpp_i64(i64 %in, i64 %old, ptr addrspace(1) %out) {
+  %tmp0 = call i64 @llvm.amdgcn.update.dpp.i64(i64 %old, i64 %in, i32 337, i32 1, i32 1, i1 0)
+  store i64 %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v2i32:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v2, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v3, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942:        v_mov_b64_dpp v[2:3], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx2 v[4:5], v[2:3], off
+
+define amdgpu_ps void @update_dpp_v2i32(<2 x i32> %in, <2 x i32> %old, ptr addrspace(1) %out) {
+  %tmp0 = call <2 x i32> @llvm.amdgcn.update.dpp.v2i32(<2 x i32> %old, <2 x i32> %in, i32 337, i32 1, i32 1, i1 0)
+  store <2 x i32> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v3i32:
+;
+; GCN-DAG:    v_mov_b32_dpp v{{[0-9]+}}, v2 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GCN-DAG:    v_mov_b32_dpp v{{[0-9]+}}, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GCN-DAG:    v_mov_b32_dpp v{{[0-9]+}}, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_dwordx3
+
+define amdgpu_ps void @update_dpp_v3i32(<3 x i32> %in, <3 x i32> %old, ptr addrspace(1) %out) {
+  %tmp0 = call <3 x i32> @llvm.amdgcn.update.dpp.v3i32(<3 x i32> %old, <3 x i32> %in, i32 337, i32 1, i32 1, i1 0)
+  store <3 x i32> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v4i32:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v6, v2 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v7, v3 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v4, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v5, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942-DAG:    v_mov_b64_dpp v[6:7], v[2:3] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX942-DAG:    v_mov_b64_dpp v[4:5], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx4 v[8:9], v[4:7], off
+
+define amdgpu_ps void @update_dpp_v4i32(<4 x i32> %in, <4 x i32> %old, ptr addrspace(1) %out) {
+  %tmp0 = call <4 x i32> @llvm.amdgcn.update.dpp.v4i32(<4 x i32> %old, <4 x i32> %in, i32 337, i32 1, i32 1, i1 0)
+  store <4 x i32> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v2i32_poison:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v1, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942:        v_mov_b64_dpp v[0:1], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx2 v[2:3], v[0:1], off
+
+define amdgpu_ps void @update_dpp_v2i32_poison(<2 x i32> %in, ptr addrspace(1) %out) {
+  %tmp0 = call <2 x i32> @llvm.amdgcn.update.dpp.v2i32(<2 x i32> poison, <2 x i32> %in, i32 337, i32 1, i32 1, i1 0)
+  store <2 x i32> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_float:
+;
+; GCN:        v_mov_b32_dpp v1, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_dword v[2:3], v1, off
+
+define amdgpu_ps void @update_dpp_float(float %in, float %old, ptr addrspace(1) %out) {
+  %tmp0 = call float @llvm.amdgcn.update.dpp.f32(float %old, float %in, i32 337, i32 1, i32 1, i1 0)
+  store float %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_double:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v2, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v3, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942:        v_mov_b64_dpp v[2:3], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx2 v[4:5], v[2:3], off
+
+define amdgpu_ps void @update_dpp_double(double %in, double %old, ptr addrspace(1) %out) {
+  %tmp0 = call double @llvm.amdgcn.update.dpp.f64(double %old, double %in, i32 337, i32 1, i32 1, i1 0)
+  store double %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_float_poison:
+;
+; GCN:        v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_dword v[{{[0-9:]+}}], v0, off
+
+define amdgpu_ps void @update_dpp_float_poison(float %in, ptr addrspace(1) %out) {
+  %tmp0 = call float @llvm.amdgcn.update.dpp.f32(float poison, float %in, i32 337, i32 1, i32 1, i1 0)
+  store float %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_double_poison:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v1, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942:        v_mov_b64_dpp v[0:1], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx2 v[2:3], v[0:1], off
+
+define amdgpu_ps void @update_dpp_double_poison(double %in, ptr addrspace(1) %out) {
+  %tmp0 = call double @llvm.amdgcn.update.dpp.f64(double poison, double %in, i32 337, i32 1, i32 1, i1 0)
+  store double %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v2f32_poison:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v1, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942:        v_mov_b64_dpp v[0:1], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx2 v[2:3], v[0:1], off
+
+define amdgpu_ps void @update_dpp_v2f32_poison(<2 x float> %in, ptr addrspace(1) %out) {
+  %tmp0 = call <2 x float> @llvm.amdgcn.update.dpp.v2f32(<2 x float> poison, <2 x float> %in, i32 337, i32 1, i32 1, i1 0)
+  store <2 x float> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v2f16_poison:
+;
+; GCN:         v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:         global_store_dword v[{{[0-9:]+}}], v0, off
+
+define amdgpu_ps void @update_dpp_v2f16_poison(<2 x half> %in, ptr addrspace(1) %out) {
+  %tmp0 = call <2 x half> @llvm.amdgcn.update.dpp.v2f16(<2 x half> poison, <2 x half> %in, i32 337, i32 1, i32 1, i1 0)
+  store <2 x half> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_v8f16_poison:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v1, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v2, v2 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v3, v3 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942-DAG:    v_mov_b64_dpp v[2:3], v[2:3] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX942-DAG:    v_mov_b64_dpp v[0:1], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:           global_store_dwordx4 v[4:5], v[0:3], off
+
+define amdgpu_ps void @update_dpp_v8f16_poison(<8 x half> %in, ptr addrspace(1) %out) {
+  %tmp0 = call <8 x half> @llvm.amdgcn.update.dpp.v8f16(<8 x half> poison, <8 x half> %in, i32 337, i32 1, i32 1, i1 0)
+  store <8 x half> %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_p3_poison:
+;
+; GCN:        v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_dword v[{{[0-9:]+}}], v0, off
+
+define amdgpu_ps void @update_dpp_p3_poison(ptr addrspace(3) %in, ptr addrspace(1) %out) {
+  %tmp0 = call ptr addrspace(3) @llvm.amdgcn.update.dpp.p3(ptr addrspace(3) poison, ptr addrspace(3) %in, i32 337, i32 1, i32 1, i1 0)
+  store ptr addrspace(3) %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_p0_poison:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+; GFX90A-DAG:    v_mov_b32_dpp v1, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GFX942-DAG:    v_mov_b64_dpp v[0:1], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:            global_store_dwordx2 v[2:3], v[0:1], off
+
+define amdgpu_ps void @update_dpp_p0_poison(ptr %in, ptr addrspace(1) %out) {
+  %tmp0 = call ptr @llvm.amdgcn.update.dpp.p0(ptr poison, ptr %in, i32 337, i32 1, i32 1, i1 0)
+  store ptr %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_i64_unsupported_dpp64_op:
+;
+; GCN-DAG:    v_mov_b32_dpp v3, v1 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+; GCN-DAG:    v_mov_b32_dpp v2, v0 quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_dwordx2 v[4:5], v[2:3], off
+
+define amdgpu_ps void @update_dpp_i64_unsupported_dpp64_op(i64 %in, i64 %old, ptr addrspace(1) %out) {
+  %tmp0 = call i64 @llvm.amdgcn.update.dpp.i64(i64 %old, i64 %in, i32 1, i32 1, i32 1, i1 0)
+  store i64 %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_i16:
+;
+; GCN:        v_mov_b32_dpp v1, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_short v[2:3], v1, off
+
+define amdgpu_ps void @update_dpp_i16(i16 %in, i16 %old, ptr addrspace(1) %out) {
+  %tmp0 = call i16 @llvm.amdgcn.update.dpp.i16(i16 %old, i16 %in, i32 337, i32 1, i32 1, i1 0)
+  store i16 %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_half:
+;
+; GCN-DAG:    v_mov_b32_dpp v1, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_short v[2:3], v1, off
+
+define amdgpu_ps void @update_dpp_half(half %in, half %old, ptr addrspace(1) %out) {
+  %tmp0 = call half @llvm.amdgcn.update.dpp.f16(half %old, half %in, i32 337, i32 1, i32 1, i1 0)
+  store half %tmp0, ptr addrspace(1) %out
+  ret void
+}
+
+; GCN-LABEL: update_dpp_bfloat:
+;
+; GCN-DAG:    v_mov_b32_dpp v1, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x1
+;
+; GCN:        global_store_short v[2:3], v1, off
+
+define amdgpu_ps void @update_dpp_bfloat(bfloat %in, bfloat %old, ptr addrspace(1) %out) {
+  %tmp0 = call bfloat @llvm.amdgcn.update.dpp.bf16(bfloat %old, bfloat %in, i32 337, i32 1, i32 1, i1 0)
+  store bfloat %tmp0, ptr addrspace(1) %out
+  ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
index a9a710f9b6723c..a7424831ae5dbf 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll
@@ -57,8 +57,10 @@ bb:
 
 ; GCN-LABEL: {{^}}update_dppi64_test:
 ; GCN:     load_{{dwordx2|b64}} v[[[SRC_LO:[0-9]+]]:[[SRC_HI:[0-9]+]]]
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
 define amdgpu_kernel void @update_dppi64_test(ptr addrspace(1) %arg, i64 %in1, i64 %in2) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds i64, ptr addrspace(1) %arg, i32 %id
@@ -70,8 +72,10 @@ define amdgpu_kernel void @update_dppi64_test(ptr addrspace(1) %arg, i64 %in1, i
 
 ; GCN-LABEL: {{^}}update_dppf64_test:
 ; GCN:     load_{{dwordx2|b64}} v[[[SRC_LO:[0-9]+]]:[[SRC_HI:[0-9]+]]]
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
 define amdgpu_kernel void @update_dppf64_test(ptr addrspace(1) %arg, double %in1, double %in2) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds double, ptr addrspace(1) %arg, i32 %id
@@ -83,8 +87,10 @@ define amdgpu_kernel void @update_dppf64_test(ptr addrspace(1) %arg, double %in1
 
 ; GCN-LABEL: {{^}}update_dppv2i32_test:
 ; GCN:     load_{{dwordx2|b64}} v[[[SRC_LO:[0-9]+]]:[[SRC_HI:[0-9]+]]]
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
 define amdgpu_kernel void @update_dppv2i32_test(ptr addrspace(1) %arg, <2 x i32> %in1, <2 x i32> %in2) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds <2 x i32>, ptr addrspace(1) %arg, i32 %id
@@ -96,8 +102,10 @@ define amdgpu_kernel void @update_dppv2i32_test(ptr addrspace(1) %arg, <2 x i32>
 
 ; GCN-LABEL: {{^}}update_dppv2f32_test:
 ; GCN:     load_{{dwordx2|b64}} v[[[SRC_LO:[0-9]+]]:[[SRC_HI:[0-9]+]]]
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
 define amdgpu_kernel void @update_dppv2f32_test(ptr addrspace(1) %arg, <2 x float> %in1, <2 x float> %in2) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds <2 x float>, ptr addrspace(1) %arg, i32 %id
@@ -109,8 +117,10 @@ define amdgpu_kernel void @update_dppv2f32_test(ptr addrspace(1) %arg, <2 x floa
 
 ; GCN-LABEL: {{^}}update_dpp_p0_test:
 ; GCN:     load_{{dwordx2|b64}} v[[[SRC_LO:[0-9]+]]:[[SRC_HI:[0-9]+]]]
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
-; GCN-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_LO]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-OPT-DAG: v_mov_b32_dpp v{{[0-9]+}}, v[[SRC_HI]] quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
+; GCN-NOOPT: v_mov_b32_dpp v{{[0-9]+}}, v{{[0-9]+}} quad_perm:[1,0,0,0] row_mask:0x1 bank_mask:0x1{{$}}
 define amdgpu_kernel void @update_dpp_p0_test(ptr addrspace(1) %arg, ptr %in1, ptr %in2) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds ptr, ptr addrspace(1) %arg, i32 %id

>From 2aec7c42262439ad6af7defcb98a9e9ccfc063c5 Mon Sep 17 00:00:00 2001
From: Stanislav Mekhanoshin <Stanislav.Mekhanoshin at amd.com>
Date: Mon, 4 Nov 2024 15:50:01 -0800
Subject: [PATCH 2/4] Formatting NFC.

---
 .../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp | 20 +++++++++----------
 llvm/lib/Target/AMDGPU/SIISelLowering.cpp     | 11 +++++-----
 2 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 23dbf50a0480b8..2fe6c4b965b95d 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -5521,12 +5521,11 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
   LLT Ty = MRI.getType(DstReg);
   unsigned Size = Ty.getSizeInBits();
 
-  unsigned SplitSize =
-      (IID == Intrinsic::amdgcn_update_dpp && (Size % 64 == 0) &&
-       ST.hasDPALU_DPP() &&
-       AMDGPU::isLegalDPALU_DPPControl(MI.getOperand(4).getImm()))
-          ? 64
-          : 32;
+  unsigned SplitSize = 32;
+  if (IID == Intrinsic::amdgcn_update_dpp && (Size % 64 == 0) &&
+      ST.hasDPALU_DPP() &&
+      AMDGPU::isLegalDPALU_DPPControl(MI.getOperand(4).getImm()))
+    SplitSize = 64;
 
   if (Size == SplitSize) {
     // Already legal
@@ -5555,11 +5554,12 @@ bool AMDGPULegalizerInfo::legalizeLaneOp(LegalizerHelper &Helper,
   if (Ty.isVector()) {
     LLT EltTy = Ty.getElementType();
     unsigned EltSize = EltTy.getSizeInBits();
-    if (EltSize == SplitSize)
+    if (EltSize == SplitSize) {
       PartialResTy = EltTy;
-    else if (EltSize == 16 || EltSize == 32)
-      PartialResTy =
-          Ty.changeElementCount(ElementCount::getFixed(SplitSize / EltSize));
+    } else if (EltSize == 16 || EltSize == 32) {
+      unsigned NElem = SplitSize / EltSize;
+      PartialResTy = Ty.changeElementCount(ElementCount::getFixed(NElem));
+    }
     // Handle all other cases via S32/S64 pieces;
   }
 
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 7aac584750d20f..eac5d04a5df551 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -6164,12 +6164,11 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
   MVT IntVT = MVT::getIntegerVT(ValSize);
   const GCNSubtarget &ST =
       DAG.getMachineFunction().getSubtarget<GCNSubtarget>();
-  unsigned SplitSize =
-      (IID == Intrinsic::amdgcn_update_dpp && (ValSize % 64 == 0) &&
-       ST.hasDPALU_DPP() &&
-       AMDGPU::isLegalDPALU_DPPControl(N->getConstantOperandVal(3)))
-          ? 64
-          : 32;
+  unsigned SplitSize = 32;
+  if (IID == Intrinsic::amdgcn_update_dpp && (ValSize % 64 == 0) &&
+      ST.hasDPALU_DPP() &&
+      AMDGPU::isLegalDPALU_DPPControl(N->getConstantOperandVal(3)))
+    SplitSize = 64;
 
   auto createLaneOp = [&DAG, &SL, N, IID](SDValue Src0, SDValue Src1,
                                           SDValue Src2, MVT ValT) -> SDValue {

>From 6f2cb141c1de746d8939f1dd1b4bf4293cc4dede Mon Sep 17 00:00:00 2001
From: Stanislav Mekhanoshin <Stanislav.Mekhanoshin at amd.com>
Date: Mon, 4 Nov 2024 16:41:28 -0800
Subject: [PATCH 3/4] Added update_dpp_v2p0_poison test

---
 .../AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll    | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll
index 841ec6efe8975d..a3d789c1ccc36f 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.gfx90a.ll
@@ -256,3 +256,21 @@ define amdgpu_ps void @update_dpp_bfloat(bfloat %in, bfloat %old, ptr addrspace(
   store bfloat %tmp0, ptr addrspace(1) %out
   ret void
 }
+
+; GCN-LABEL: update_dpp_v2p0_poison:
+;
+; GFX90A-DAG:    v_mov_b32_dpp v0, v0 row_newbcast:1 row_mask:0x1 bank_mask:0x2
+; GFX90A-DAG:    v_mov_b32_dpp v1, v1 row_newbcast:1 row_mask:0x1 bank_mask:0x2
+; GFX90A-DAG:    v_mov_b32_dpp v2, v2 row_newbcast:1 row_mask:0x1 bank_mask:0x2
+; GFX90A-DAG:    v_mov_b32_dpp v3, v3 row_newbcast:1 row_mask:0x1 bank_mask:0x2
+;
+; GFX942-DAG:    v_mov_b64_dpp v[2:3], v[2:3] row_newbcast:1 row_mask:0x1 bank_mask:0x2
+; GFX942-DAG:    v_mov_b64_dpp v[0:1], v[0:1] row_newbcast:1 row_mask:0x1 bank_mask:0x2
+;
+; GCN:           global_store_dwordx4 v[4:5], v[0:3], off
+
+define amdgpu_ps void @update_dpp_v2p0_poison(<2 x ptr> %in, ptr addrspace(1) %out) {
+  %tmp0 = call <2 x ptr> @llvm.amdgcn.update.dpp.v2p0(<2 x ptr> poison, <2 x ptr> %in, i32 337, i32 1, i32 2, i1 0)
+  store <2 x ptr> %tmp0, ptr addrspace(1) %out
+  ret void
+}

>From 2c5bf20e227f674cf1d26f4f6f9420ce77ad8584 Mon Sep 17 00:00:00 2001
From: Stanislav Mekhanoshin <Stanislav.Mekhanoshin at amd.com>
Date: Mon, 4 Nov 2024 22:45:08 -0800
Subject: [PATCH 4/4] Changed to TLI.getSubtarget()

---
 llvm/lib/Target/AMDGPU/SIISelLowering.cpp | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index eac5d04a5df551..ae47a89864fbb8 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -6162,11 +6162,10 @@ static SDValue lowerLaneOp(const SITargetLowering &TLI, SDNode *N,
                        IID == Intrinsic::amdgcn_set_inactive_chain_arg;
   SDLoc SL(N);
   MVT IntVT = MVT::getIntegerVT(ValSize);
-  const GCNSubtarget &ST =
-      DAG.getMachineFunction().getSubtarget<GCNSubtarget>();
+  const GCNSubtarget *ST = TLI.getSubtarget();
   unsigned SplitSize = 32;
   if (IID == Intrinsic::amdgcn_update_dpp && (ValSize % 64 == 0) &&
-      ST.hasDPALU_DPP() &&
+      ST->hasDPALU_DPP() &&
       AMDGPU::isLegalDPALU_DPPControl(N->getConstantOperandVal(3)))
     SplitSize = 64;
 



More information about the llvm-commits mailing list