AMDGPU: Fix multi-use shl/add combine

[arsenm]

Summary

This was using a custom function that didn't handle the
addressing modes properly for private. Use
isLegalAddressingMode to avoid duplicating this.

Additionally, skip the combine if there is only one use
since the standard combine will handle it.

[rampitec]

LGTM

[arsenm]

r318013

lib/Target/AMDGPU/SIISelLowering.h

class SITargetLowering final : public AMDGPUTargetLowering {
   SDValue lowerTRAP(SDValue Op, SelectionDAG &DAG) const;
 
   void adjustWritemask(MachineSDNode *&N, SelectionDAG &DAG) const;
 
   SDValue performUCharToFloatCombine(SDNode *N,
                                      DAGCombinerInfo &DCI) const;
   SDValue performSHLPtrCombine(SDNode *N,
                                unsigned AS,
+                               EVT MemVT,
                                DAGCombinerInfo &DCI) const;
 
   SDValue performMemSDNodeCombine(MemSDNode *N, DAGCombinerInfo &DCI) const;
 
   SDValue splitBinaryBitConstantOp(DAGCombinerInfo &DCI, const SDLoc &SL,
                                    unsigned Opc, SDValue LHS,
                                    const ConstantSDNode *CRHS) const;
 

lib/Target/AMDGPU/SIISelLowering.cpp

if (DAG.MaskedValueIsZero(Src, APInt::getHighBitsSet(32, 24))) {
       DCI.AddToWorklist(Cvt.getNode());
       return Cvt;
     }
   }
 
   return SDValue();
 }
 
-/// \brief Return true if the given offset Size in bytes can be folded into
-/// the immediate offsets of a memory instruction for the given address space.
-static bool canFoldOffset(unsigned OffsetSize, unsigned AS,
-                          const SISubtarget &STI) {
-  auto AMDGPUASI = STI.getAMDGPUAS();
-  if (AS == AMDGPUASI.GLOBAL_ADDRESS || AS == AMDGPUASI.PRIVATE_ADDRESS) {
-    // MUBUF instructions a 12-bit offset in bytes.
-    return isUInt<12>(OffsetSize);
-  }
-  if (AS == AMDGPUASI.CONSTANT_ADDRESS) {
-    // SMRD instructions have an 8-bit offset in dwords on SI and
-    // a 20-bit offset in bytes on VI.
-    if (STI.getGeneration() >= SISubtarget::VOLCANIC_ISLANDS)
-      return isUInt<20>(OffsetSize);
-    else
-      return (OffsetSize % 4 == 0) && isUInt<8>(OffsetSize / 4);
-  }
-  if (AS == AMDGPUASI.LOCAL_ADDRESS ||
-      AS == AMDGPUASI.REGION_ADDRESS) {
-    // The single offset versions have a 16-bit offset in bytes.
-    return isUInt<16>(OffsetSize);
-  }
-
-  return false;
-}
-
 // (shl (add x, c1), c2) -> add (shl x, c2), (shl c1, c2)
 
 // This is a variant of
 // (mul (add x, c1), c2) -> add (mul x, c2), (mul c1, c2),
 //
 // The normal DAG combiner will do this, but only if the add has one use since
 // that would increase the number of instructions.
 //
 // This prevents us from seeing a constant offset that can be folded into a
 // memory instruction's addressing mode. If we know the resulting add offset of
 // a pointer can be folded into an addressing offset, we can replace the pointer
 // operand with the add of new constant offset. This eliminates one of the uses,
 // and may allow the remaining use to also be simplified.
 //
 SDValue SITargetLowering::performSHLPtrCombine(SDNode *N,
                                                unsigned AddrSpace,
+                                               EVT MemVT,
                                                DAGCombinerInfo &DCI) const {
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);
 
-  if (N0.getOpcode() != ISD::ADD)
+  // We only do this to handle cases where it's profitable when there are
+  // multiple uses of the add, so defer to the standard combine.
+  // TODO: Support or
+  if (N0.getOpcode() != ISD::ADD || N0->hasOneUse())
     return SDValue();
 
   const ConstantSDNode *CN1 = dyn_cast<ConstantSDNode>(N1);
   if (!CN1)
     return SDValue();
 
   const ConstantSDNode *CAdd = dyn_cast<ConstantSDNode>(N0.getOperand(1));
   if (!CAdd)
     return SDValue();
 
   // If the resulting offset is too large, we can't fold it into the addressing
   // mode offset.
   APInt Offset = CAdd->getAPIntValue() << CN1->getAPIntValue();
-  if (!canFoldOffset(Offset.getZExtValue(), AddrSpace, *getSubtarget()))
+  Type *Ty = MemVT.getTypeForEVT(*DCI.DAG.getContext());
+
+  AddrMode AM;
+  AM.HasBaseReg = true;
+  AM.BaseOffs = Offset.getSExtValue();
+  if (!isLegalAddressingMode(DCI.DAG.getDataLayout(), AM, Ty, AddrSpace))
     return SDValue();
 
   SelectionDAG &DAG = DCI.DAG;
   SDLoc SL(N);
   EVT VT = N->getValueType(0);
 
   SDValue ShlX = DAG.getNode(ISD::SHL, SL, VT, N0.getOperand(0), N1);
   SDValue COffset = DAG.getConstant(Offset, SL, MVT::i32);
 
-  llvm_unreachable("this happens");
-
   // XXX - Does this preserve NUW?
   return DAG.getNode(ISD::ADD, SL, VT, ShlX, COffset);
 }
 
 SDValue SITargetLowering::performMemSDNodeCombine(MemSDNode *N,
                                                   DAGCombinerInfo &DCI) const {
   SDValue Ptr = N->getBasePtr();
   SelectionDAG &DAG = DCI.DAG;
   SDLoc SL(N);
 
   // TODO: We could also do this for multiplies.
-  unsigned AS = N->getAddressSpace();
   if (Ptr.getOpcode() == ISD::SHL) {
-    SDValue NewPtr = performSHLPtrCombine(Ptr.getNode(), AS, DCI);
+    SDValue NewPtr = performSHLPtrCombine(Ptr.getNode(),  N->getAddressSpace(),
+                                          N->getMemoryVT(), DCI);
     if (NewPtr) {
       SmallVector<SDValue, 8> NewOps(N->op_begin(), N->op_end());
 
       NewOps[N->getOpcode() == ISD::STORE ? 2 : 1] = NewPtr;
       return SDValue(DAG.UpdateNodeOperands(N, NewOps), 0);
     }
   }
 

test/CodeGen/AMDGPU/shl_add_ptr.ll

-; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -mattr=+load-store-opt -enable-misched < %s | FileCheck -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs -mattr=+load-store-opt -enable-misched < %s | FileCheck -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
 
 ; Test that doing a shift of a pointer with a constant add will be
 ; folded into the constant offset addressing mode even if the add has
 ; multiple uses. This is relevant to accessing 2 separate, adjacent
 ; LDS globals.
 
 
 declare i32 @llvm.amdgcn.workitem.id.x() #1
 
 @lds0 = addrspace(3) global [512 x float] undef, align 4
 @lds1 = addrspace(3) global [512 x float] undef, align 4
 
 
 ; Make sure the (add tid, 2) << 2 gets folded into the ds's offset as (tid << 2) + 8
 
-; SI-LABEL: {{^}}load_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_read_b32 {{v[0-9]+}}, [[PTR]] offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}load_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_read_b32 {{v[0-9]+}}, [[PTR]] offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @load_shl_base_lds_0(float addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds0, i32 0, i32 %idx.0
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   store float %val0, float addrspace(1)* %out
   ret void
 }
 
 ; Make sure once the first use is folded into the addressing mode, the
 ; remaining add use goes through the normal shl + add constant fold.
 
-; SI-LABEL: {{^}}load_shl_base_lds_1:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_read_b32 [[RESULT:v[0-9]+]], [[PTR]] offset:8
-; SI: v_add_i32_e32 [[ADDUSE:v[0-9]+]], vcc, 8, v{{[0-9]+}}
-; SI-DAG: buffer_store_dword [[RESULT]]
-; SI-DAG: buffer_store_dword [[ADDUSE]]
-; SI: s_endpgm
+; GCN-LABEL: {{^}}load_shl_base_lds_1:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_read_b32 [[RESULT:v[0-9]+]], [[PTR]] offset:8
+; GCN: v_add_i32_e32 [[ADDUSE:v[0-9]+]], vcc, 8, v{{[0-9]+}}
+; GCN-DAG: buffer_store_dword [[RESULT]]
+; GCN-DAG: buffer_store_dword [[ADDUSE]]
+; GCN: s_endpgm
 define amdgpu_kernel void @load_shl_base_lds_1(float addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds0, i32 0, i32 %idx.0
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   %shl_add_use = shl i32 %idx.0, 2
   store i32 %shl_add_use, i32 addrspace(1)* %add_use, align 4
   store float %val0, float addrspace(1)* %out
   ret void
 }
 
 @maxlds = addrspace(3) global [65536 x i8] undef, align 4
 
-; SI-LABEL: {{^}}load_shl_base_lds_max_offset
-; SI: ds_read_u8 v{{[0-9]+}}, v{{[0-9]+}} offset:65535
-; SI: s_endpgm
+; GCN-LABEL: {{^}}load_shl_base_lds_max_offset
+; GCN: ds_read_u8 v{{[0-9]+}}, v{{[0-9]+}} offset:65535
+; GCN: s_endpgm
 define amdgpu_kernel void @load_shl_base_lds_max_offset(i8 addrspace(1)* %out, i8 addrspace(3)* %lds, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 65535
   %arrayidx0 = getelementptr inbounds [65536 x i8], [65536 x i8] addrspace(3)* @maxlds, i32 0, i32 %idx.0
   %val0 = load i8, i8 addrspace(3)* %arrayidx0
   store i32 %idx.0, i32 addrspace(1)* %add_use
   store i8 %val0, i8 addrspace(1)* %out
   ret void
 }
 
 ; The two globals are placed adjacent in memory, so the same base
 ; pointer can be used with an offset into the second one.
 
-; SI-LABEL: {{^}}load_shl_base_lds_2:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: s_mov_b32 m0, -1
-; SI-NEXT: ds_read2st64_b32 {{v\[[0-9]+:[0-9]+\]}}, [[PTR]] offset0:1 offset1:9
-; SI: s_endpgm
+; GCN-LABEL: {{^}}load_shl_base_lds_2:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: s_mov_b32 m0, -1
+; GCN-NEXT: ds_read2st64_b32 {{v\[[0-9]+:[0-9]+\]}}, [[PTR]] offset0:1 offset1:9
+; GCN: s_endpgm
 define amdgpu_kernel void @load_shl_base_lds_2(float addrspace(1)* %out) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 64
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds0, i32 0, i32 %idx.0
   %val0 = load float, float addrspace(3)* %arrayidx0, align 4
   %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds1, i32 0, i32 %idx.0
   %val1 = load float, float addrspace(3)* %arrayidx1, align 4
   %sum = fadd float %val0, %val1
   store float %sum, float addrspace(1)* %out, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}store_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_write_b32 [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}store_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_write_b32 [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @store_shl_base_lds_0(float addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds0, i32 0, i32 %idx.0
   store float 1.0, float addrspace(3)* %arrayidx0, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }

 ;   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
 ;   %val = load atomic i32, i32 addrspace(3)* %arrayidx0 seq_cst, align 4
 ;   store i32 %val, i32 addrspace(1)* %out, align 4
 ;   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
 ;   ret void
 ; }
 
 
-; SI-LABEL: {{^}}atomic_cmpxchg_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_cmpst_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}}, {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_cmpxchg_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_cmpst_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}}, {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_cmpxchg_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use, i32 %swap) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %pair = cmpxchg i32 addrspace(3)* %arrayidx0, i32 7, i32 %swap seq_cst monotonic
   %result = extractvalue { i32, i1 } %pair, 0
   store i32 %result, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_swap_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_wrxchg_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_swap_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_wrxchg_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_swap_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw xchg i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_add_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_add_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_add_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_add_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_add_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw add i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_sub_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_sub_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_sub_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_sub_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_sub_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw sub i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_and_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_and_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_and_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_and_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_and_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw and i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_or_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_or_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_or_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_or_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_or_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw or i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_xor_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_xor_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_xor_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_xor_rtn_b32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_xor_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw xor i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
 ; define amdgpu_kernel void @atomic_nand_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
 ;   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
 ;   %idx.0 = add nsw i32 %tid.x, 2
 ;   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
 ;   %val = atomicrmw nand i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
 ;   store i32 %val, i32 addrspace(1)* %out, align 4
 ;   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
 ;   ret void
 ; }
 
-; SI-LABEL: {{^}}atomic_min_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_min_rtn_i32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_min_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_min_rtn_i32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_min_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw min i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_max_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_max_rtn_i32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_max_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_max_rtn_i32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_max_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw max i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_umin_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_min_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_umin_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_min_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_umin_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw umin i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
-; SI-LABEL: {{^}}atomic_umax_shl_base_lds_0:
-; SI: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
-; SI: ds_max_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
-; SI: s_endpgm
+; GCN-LABEL: {{^}}atomic_umax_shl_base_lds_0:
+; GCN: v_lshlrev_b32_e32 [[PTR:v[0-9]+]], 2, {{v[0-9]+}}
+; GCN: ds_max_rtn_u32 {{v[0-9]+}}, [[PTR]], {{v[0-9]+}} offset:8
+; GCN: s_endpgm
 define amdgpu_kernel void @atomic_umax_shl_base_lds_0(i32 addrspace(1)* %out, i32 addrspace(1)* %add_use) #0 {
   %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
   %idx.0 = add nsw i32 %tid.x, 2
   %arrayidx0 = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds2, i32 0, i32 %idx.0
   %val = atomicrmw umax i32 addrspace(3)* %arrayidx0, i32 3 seq_cst
   store i32 %val, i32 addrspace(1)* %out, align 4
   store i32 %idx.0, i32 addrspace(1)* %add_use, align 4
   ret void
 }
 
+; GCN-LABEL: {{^}}shl_add_ptr_combine_2use_lds:
+; GCN: v_lshlrev_b32_e32 [[SCALE0:v[0-9]+]], 3, v0
+; GCN: ds_write_b32 [[SCALE0]], v{{[0-9]+}} offset:32
+
+; GCN: v_lshlrev_b32_e32 [[SCALE1:v[0-9]+]], 4, v0
+; GCN: ds_write_b32 [[SCALE1]], v{{[0-9]+}} offset:64
+define void @shl_add_ptr_combine_2use_lds(i32 %idx) #0 {
+  %idx.add = add nuw i32 %idx, 4
+  %shl0 = shl i32 %idx.add, 3
+  %shl1 = shl i32 %idx.add, 4
+  %ptr0 = inttoptr i32 %shl0 to i32 addrspace(3)*
+  %ptr1 = inttoptr i32 %shl1 to i32 addrspace(3)*
+  store volatile i32 9, i32 addrspace(3)* %ptr0
+  store volatile i32 10, i32 addrspace(3)* %ptr1
+  ret void
+}
+
+; GCN-LABEL: {{^}}shl_add_ptr_combine_2use_max_lds_offset:
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE0:v[0-9]+]], 3, v0
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE1:v[0-9]+]], 4, v0
+; GCN-DAG: ds_write_b32 [[SCALE0]], v{{[0-9]+}} offset:65528
+; GCN-DAG: v_add_i32_e32 [[ADD1:v[0-9]+]], vcc, 0x1fff0, [[SCALE1]]
+; GCN: ds_write_b32 [[ADD1]], v{{[0-9]+$}}
+define void @shl_add_ptr_combine_2use_max_lds_offset(i32 %idx) #0 {
+  %idx.add = add nuw i32 %idx, 8191
+  %shl0 = shl i32 %idx.add, 3
+  %shl1 = shl i32 %idx.add, 4
+  %ptr0 = inttoptr i32 %shl0 to i32 addrspace(3)*
+  %ptr1 = inttoptr i32 %shl1 to i32 addrspace(3)*
+  store volatile i32 9, i32 addrspace(3)* %ptr0
+  store volatile i32 10, i32 addrspace(3)* %ptr1
+  ret void
+}
+
+; GCN-LABEL: {{^}}shl_add_ptr_combine_2use_both_max_lds_offset:
+; GCN: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 0x1000, v0
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE0:v[0-9]+]], 4, [[ADD]]
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE1:v[0-9]+]], 5, [[ADD]]
+; GCN-DAG: ds_write_b32 [[SCALE0]], v{{[0-9]+$}}
+; GCN: ds_write_b32 [[SCALE1]], v{{[0-9]+$}}
+define void @shl_add_ptr_combine_2use_both_max_lds_offset(i32 %idx) #0 {
+  %idx.add = add nuw i32 %idx, 4096
+  %shl0 = shl i32 %idx.add, 4
+  %shl1 = shl i32 %idx.add, 5
+  %ptr0 = inttoptr i32 %shl0 to i32 addrspace(3)*
+  %ptr1 = inttoptr i32 %shl1 to i32 addrspace(3)*
+  store volatile i32 9, i32 addrspace(3)* %ptr0
+  store volatile i32 10, i32 addrspace(3)* %ptr1
+  ret void
+}
+
+; GCN-LABEL: {{^}}shl_add_ptr_combine_2use_private:
+; GCN: v_lshlrev_b32_e32 [[SCALE0:v[0-9]+]], 2, v0
+; GCN: buffer_store_dword v{{[0-9]+}}, [[SCALE0]], s[0:3], s4 offen offset:16
+
+; GCN: v_lshlrev_b32_e32 [[SCALE1:v[0-9]+]], 3, v0
+; GCN: buffer_store_dword v{{[0-9]+}}, [[SCALE1]], s[0:3], s4 offen offset:32
+define void @shl_add_ptr_combine_2use_private(i16 zeroext %idx.arg) #0 {
+  %idx = zext i16 %idx.arg to i32
+  %idx.add = add nuw i32 %idx, 4
+  %shl0 = shl i32 %idx.add, 2
+  %shl1 = shl i32 %idx.add, 3
+  %ptr0 = inttoptr i32 %shl0 to i32*
+  %ptr1 = inttoptr i32 %shl1 to i32*
+  store volatile i32 9, i32* %ptr0
+  store volatile i32 10, i32* %ptr1
+  ret void
+}
+
+; GCN-LABEL: {{^}}shl_add_ptr_combine_2use_max_private_offset:
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE0:v[0-9]+]], 3, v0
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE1:v[0-9]+]], 4, v0
+; GCN-DAG: buffer_store_dword v{{[0-9]+}}, [[SCALE0]], s[0:3], s4 offen offset:4088
+; GCN-DAG: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 0x1ff0, [[SCALE1]]
+; GCN: buffer_store_dword v{{[0-9]+}}, [[ADD]], s[0:3], s4 offen{{$}}
+define void @shl_add_ptr_combine_2use_max_private_offset(i16 zeroext %idx.arg) #0 {
+  %idx = zext i16 %idx.arg to i32
+  %idx.add = add nuw i32 %idx, 511
+  %shl0 = shl i32 %idx.add, 3
+  %shl1 = shl i32 %idx.add, 4
+  %ptr0 = inttoptr i32 %shl0 to i32*
+  %ptr1 = inttoptr i32 %shl1 to i32*
+  store volatile i32 9, i32* %ptr0
+  store volatile i32 10, i32* %ptr1
+  ret void
+}
+; GCN-LABEL: {{^}}shl_add_ptr_combine_2use_both_max_private_offset:
+; GCN: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 0x100, v0
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE0:v[0-9]+]], 4, [[ADD]]
+; GCN-DAG: v_lshlrev_b32_e32 [[SCALE1:v[0-9]+]], 5, [[ADD]]
+; GCN-DAG: buffer_store_dword v{{[0-9]+}}, [[SCALE0]], s[0:3], s4 offen{{$}}
+; GCN: buffer_store_dword v{{[0-9]+}}, [[SCALE1]], s[0:3], s4 offen{{$}}
+define void @shl_add_ptr_combine_2use_both_max_private_offset(i16 zeroext %idx.arg) #0 {
+  %idx = zext i16 %idx.arg to i32
+  %idx.add = add nuw i32 %idx, 256
+  %shl0 = shl i32 %idx.add, 4
+  %shl1 = shl i32 %idx.add, 5
+  %ptr0 = inttoptr i32 %shl0 to i32*
+  %ptr1 = inttoptr i32 %shl1 to i32*
+  store volatile i32 9, i32* %ptr0
+  store volatile i32 10, i32* %ptr1
+  ret void
+}
+
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }