AMDGPU: Fix ds_{read,write}2_b64 on SI/gfx6

[nhaehnle]

Summary

Offsets for these instructions are in dwords, like for the b32 variants.

This is exposed by a combination of:

• better load / store vectorization in LLVM
• an in-flight change in Mesa to specify an increased alignment for LDS in compute shaders, which enables more vectorization opportunities

This has been observed on real hardware in the following tests:

• dEQP-GLES31.functional.compute.shared_var.basic_type.mat4_{lowp,mediump,highp}
• dEQP-GLES31.functional.compute.shared_var.work_group_size.mat4_{64_1_1, 1_64_1, 1_1_64}

Change-Id: I63c20afd1467b126199be2891ab45451b0430103

[rampitec]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

319

Was it really changed in CI?

[arsenm]

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

319

I doubt this changed, and the manual for SI and gfx9 plainly state it uses 8. This should also not use a hardcoded generation check if this is really true

[nhaehnle]

After some more investigation, you're right.

However, there is some SI-specific bug that none of D60459, D61313, and D60457 fix. It's triggered by a trivial modification of an unfortunately relatively complex shader and is extremely sensitive to code changes, so it might be some missing hazard or similar. I'll have to investigate that further.

lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

 private:
   const GCNSubtarget *STM = nullptr;
   const SIInstrInfo *TII = nullptr;
   const SIRegisterInfo *TRI = nullptr;
   MachineRegisterInfo *MRI = nullptr;
   AliasAnalysis *AA = nullptr;
   bool OptimizeAgain;
 
-  static bool offsetsCanBeCombined(CombineInfo &CI);
+  bool offsetsCanBeCombined(CombineInfo &CI);
   static bool widthsFit(const GCNSubtarget &STM, const CombineInfo &CI);
   static unsigned getNewOpcode(const CombineInfo &CI);
   static std::pair<unsigned, unsigned> getSubRegIdxs(const CombineInfo &CI);
   const TargetRegisterClass *getTargetRegisterClass(const CombineInfo &CI);
   unsigned getOpcodeWidth(const MachineInstr &MI);
   InstClassEnum getInstClass(unsigned Opc);
   unsigned getRegs(unsigned Opc);
 
bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI) {
   // be useful?
   if (CI.Offset0 == CI.Offset1)
     return false;
 
   // This won't be valid if the offset isn't aligned.
   if ((CI.Offset0 % CI.EltSize != 0) || (CI.Offset1 % CI.EltSize != 0))
     return false;
 
-  unsigned EltOffset0 = CI.Offset0 / CI.EltSize;
-  unsigned EltOffset1 = CI.Offset1 / CI.EltSize;
+  unsigned OffsetSize = CI.EltSize;
+  if ((CI.InstClass == DS_READ || CI.InstClass == DS_WRITE) &&
+      (OffsetSize == 8 && STM->getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS))
+    OffsetSize = 4;
+  unsigned EltOffset0 = CI.Offset0 / OffsetSize;
+  unsigned EltOffset1 = CI.Offset1 / OffsetSize;
   CI.UseST64 = false;
   CI.BaseOff = 0;
 
   // Handle SMEM and VMEM instructions.
   if ((CI.InstClass != DS_READ) && (CI.InstClass != DS_WRITE)) {
     return (EltOffset0 + CI.Width0 == EltOffset1 ||
             EltOffset1 + CI.Width1 == EltOffset0) &&
            CI.GLC0 == CI.GLC1 &&
if (isUInt<8>(EltOffset0) && isUInt<8>(EltOffset1)) {
     return true;
   }
 
   // Try to shift base address to decrease offsets.
   unsigned OffsetDiff = std::abs((int)EltOffset1 - (int)EltOffset0);
   CI.BaseOff = std::min(CI.Offset0, CI.Offset1);
 
   if ((OffsetDiff % 64 == 0) && isUInt<8>(OffsetDiff / 64)) {
-    CI.Offset0 = (EltOffset0 - CI.BaseOff / CI.EltSize) / 64;
-    CI.Offset1 = (EltOffset1 - CI.BaseOff / CI.EltSize) / 64;
+    CI.Offset0 = (EltOffset0 - CI.BaseOff / OffsetSize) / 64;
+    CI.Offset1 = (EltOffset1 - CI.BaseOff / OffsetSize) / 64;
     CI.UseST64 = true;
     return true;
   }
 
   if (isUInt<8>(OffsetDiff)) {
-    CI.Offset0 = EltOffset0 - CI.BaseOff / CI.EltSize;
-    CI.Offset1 = EltOffset1 - CI.BaseOff / CI.EltSize;
+    CI.Offset0 = EltOffset0 - CI.BaseOff / OffsetSize;
+    CI.Offset1 = EltOffset1 - CI.BaseOff / OffsetSize;
     return true;
   }
 
   return false;
 }
 
 bool SILoadStoreOptimizer::widthsFit(const GCNSubtarget &STM,
                                      const CombineInfo &CI) {

test/CodeGen/AMDGPU/load-local-i16.ll

 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,SI,SICIVI,FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,SICIVI,GFX89,FUNC %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,GFX9,GFX89,FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,SICIVI,GFX89,CIPLUS,FUNC %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,GFX9,GFX89,CIPLUS,FUNC %s
 ; RUN: llc -march=r600 -mcpu=redwood -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=EG -check-prefix=FUNC %s
 
 ; Testing for ds_read/write_b128
 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=+enable-ds128 < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=CIVI,FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+enable-ds128 < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=CIVI,FUNC %s
 
 ; FUNC-LABEL: {{^}}local_load_i16:
 ; GFX9-NOT: m0
entry:
   store <4 x i16> %ld, <4 x i16> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_load_v8i16:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; SI: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; CIPLUS: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 define amdgpu_kernel void @local_load_v8i16(<8 x i16> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) {
 entry:
   %ld = load <8 x i16>, <8 x i16> addrspace(3)* %in
   store <8 x i16> %ld, <8 x i16> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_load_v16i16:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
 
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
define amdgpu_kernel void @local_sextload_v4i16_to_v4i32(<4 x i32> addrspace(3)* %out, <4 x i16> addrspace(3)* %in) #0 {
   store <4 x i32> %ext, <4 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_zextload_v8i16_to_v8i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; SI: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; CIPLUS: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 define amdgpu_kernel void @local_zextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) #0 {
   %load = load <8 x i16>, <8 x i16> addrspace(3)* %in
   %ext = zext <8 x i16> %load to <8 x i32>
   store <8 x i32> %ext, <8 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_sextload_v8i16_to_v8i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; SI: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; CIPLUS: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 ; EG-DAG: BFE_INT
 define amdgpu_kernel void @local_sextload_v8i16_to_v8i32(<8 x i32> addrspace(3)* %out, <8 x i16> addrspace(3)* %in) #0 {
   %load = load <8 x i16>, <8 x i16> addrspace(3)* %in
   %ext = sext <8 x i16> %load to <8 x i32>
   store <8 x i32> %ext, <8 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_zextload_v16i16_to_v16i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6{{$}}
+
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
 
 ; GCN: ds_write2_b64
 ; GCN: ds_write2_b64
 ; GCN: ds_write2_b64
 ; GCN: ds_write2_b64
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
define amdgpu_kernel void @local_zextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x i16> addrspace(3)* %in) #0 {
   store <16 x i32> %ext, <16 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_sextload_v16i16_to_v16i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6{{$}}
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
define amdgpu_kernel void @local_sextload_v16i16_to_v16i32(<16 x i32> addrspace(3)* %out, <16 x i16> addrspace(3)* %in) #0 {
   store <16 x i32> %ext, <16 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_zextload_v32i16_to_v32i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:8 offset1:10
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:14
+
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
define amdgpu_kernel void @local_zextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x i16> addrspace(3)* %in) #0 {
   store <32 x i32> %ext, <32 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_sextload_v32i16_to_v32i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:13
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:10 offset1:11
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:9
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:6 offset1:7
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:5
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:1
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:8 offset1:10
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:14
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:28 offset1:30
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:24 offset1:26
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:20 offset1:22
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:16 offset1:18
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:14
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:10
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:6
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:2
+
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:13
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:10 offset1:11
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:9
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:6 offset1:7
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:5
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:1
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
define amdgpu_kernel void @local_sextload_v32i16_to_v32i32(<32 x i32> addrspace(3)* %out, <32 x i16> addrspace(3)* %in) #0 {
   store <32 x i32> %ext, <32 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_zextload_v64i16_to_v64i32:
 ; GFX9-NOT: m0
 ; SICIVI: s_mov_b32 m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:14 offset1:15
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:8 offset1:9
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:13
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:10 offset1:11
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:30 offset1:31
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:28 offset1:29
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:26 offset1:27
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:24 offset1:25
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:22 offset1:23
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:20 offset1:21
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:18 offset1:19
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:16 offset1:17
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:13
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:10 offset1:11
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:9
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:6 offset1:7
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:5
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:1
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:28 offset1:30
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:8 offset1:10
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:14
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:16 offset1:18
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:24 offset1:26
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:20 offset1:22
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:60 offset1:62
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:56 offset1:58
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:52 offset1:54
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:48 offset1:50
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:44 offset1:46
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:40 offset1:42
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:36 offset1:38
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:32 offset1:34
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:28 offset1:30
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:24 offset1:26
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:20 offset1:22
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:16 offset1:18
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:14
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:10
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:6
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:2
+
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:14 offset1:15
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:8 offset1:9
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:13
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:10 offset1:11
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:30 offset1:31
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:28 offset1:29
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:26 offset1:27
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:24 offset1:25
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:22 offset1:23
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:20 offset1:21
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:18 offset1:19
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:16 offset1:17
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:13
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:10 offset1:11
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:9
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:6 offset1:7
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:5
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:1
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET

test/CodeGen/AMDGPU/load-local-i32.ll

 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,FUNC %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,SICIVI,CIPLUS,FUNC %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,VI,CIPLUS,FUNC %s
 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
 ; Testing for ds_read/write_128
 ; RUN: llc -march=amdgcn -mcpu=tahiti -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=SI,FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+enable-ds128 < %s | FileCheck -check-prefixes=CIVI,FUNC %s
 
 ; FUNC-LABEL: {{^}}local_load_i32:
entry:
   store <3 x i32> %ld, <3 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_load_v4i32:
 ; SICIVI: s_mov_b32 m0, -1
 ; GFX9-NOT: m0
 
-; GCN: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; SI: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; CIPLUS: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
 
 define amdgpu_kernel void @local_load_v4i32(<4 x i32> addrspace(3)* %out, <4 x i32> addrspace(3)* %in) #0 {
 entry:
   %ld = load <4 x i32>, <4 x i32> addrspace(3)* %in
   store <4 x i32> %ld, <4 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_load_v8i32:
 ; SICIVI: s_mov_b32 m0, -1
 ; GFX9-NOT: m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
 define amdgpu_kernel void @local_load_v8i32(<8 x i32> addrspace(3)* %out, <8 x i32> addrspace(3)* %in) #0 {
 entry:
   %ld = load <8 x i32>, <8 x i32> addrspace(3)* %in
   store <8 x i32> %ld, <8 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_load_v16i32:
 ; SICIVI: s_mov_b32 m0, -1
 ; GFX9-NOT: m0
 
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
-; GCN-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:6 offset1:7
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:5
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:1
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:12 offset1:14{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:8 offset1:10{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:6{{$}}
+; SI-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:12 offset1:14
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:8 offset1:10
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:6
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:2
+
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:6 offset1:7{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:4 offset1:5{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset0:2 offset1:3{{$}}
+; CIPLUS-DAG: ds_read2_b64 v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:6 offset1:7
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:4 offset1:5
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}} offset1:1
 define amdgpu_kernel void @local_load_v16i32(<16 x i32> addrspace(3)* %out, <16 x i32> addrspace(3)* %in) #0 {
 entry:
   %ld = load <16 x i32>, <16 x i32> addrspace(3)* %in
   store <16 x i32> %ld, <16 x i32> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_zextload_i32_to_i64:

test/CodeGen/AMDGPU/load-local-i8.ll

 ; RUN: llc -march=amdgcn -mtriple=amdgcn-- -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,SI,SICIVI,FUNC %s
-; RUN: llc -march=amdgcn -mtriple=amdgcn-- -mcpu=tonga -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,VI,SICIVI,FUNC %s
-; RUN: llc -march=amdgcn -mtriple=amdgcn-- -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,GFX9,FUNC %s
+; RUN: llc -march=amdgcn -mtriple=amdgcn-- -mcpu=tonga -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,VI,SICIVI,CIPLUS,FUNC %s
+; RUN: llc -march=amdgcn -mtriple=amdgcn-- -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=GCN,GFX9,CIPLUS,FUNC %s
 ; RUN: llc -march=r600 -mtriple=r600-- -mcpu=redwood -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=EG -check-prefix=FUNC %s
 
 ; Testing for ds_read/write_b128
 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=+enable-ds128 < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=CIVI,FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+enable-ds128 < %s | FileCheck -allow-deprecated-dag-overlap -check-prefixes=CIVI,FUNC %s
 
 ; FUNC-LABEL: {{^}}local_load_i8:
 ; GCN-NOT: s_wqm_b64

 entry:
   %ld = load <8 x i8>, <8 x i8> addrspace(3)* %in
   store <8 x i8> %ld, <8 x i8> addrspace(3)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}local_load_v16i8:
 ; GFX9-NOT: m0
-; GCN: ds_read2_b64  v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, v{{[0-9]+}} offset1:1{{$}}
-; GCN: ds_write2_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:{{[0-9]+}}], v[{{[0-9]+}}:[[HI]]{{\]}} offset1:1{{$}}
+; SI: ds_read2_b64  v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, v{{[0-9]+}} offset1:2{{$}}
+; SI: ds_write2_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:{{[0-9]+}}], v[{{[0-9]+}}:[[HI]]{{\]}} offset1:2{{$}}
+; CIPLUS: ds_read2_b64  v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, v{{[0-9]+}} offset1:1{{$}}
+; CIPLUS: ds_write2_b64 v{{[0-9]+}}, v{{\[}}[[LO]]:{{[0-9]+}}], v[{{[0-9]+}}:[[HI]]{{\]}} offset1:1{{$}}
 
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 ; EG: LDS_READ_RET
 define amdgpu_kernel void @local_load_v16i8(<16 x i8> addrspace(3)* %out, <16 x i8> addrspace(3)* %in) #0 {
 entry:
   %ld = load <16 x i8>, <16 x i8> addrspace(3)* %in

test/CodeGen/AMDGPU/local-64.ll

 ; RUN: llc -march=amdgcn -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI,SICIVI %s
-; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI,SICIVI %s
-; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,SICIVI %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9 %s
+; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI,SICIVI,CIPLUS %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,SICIVI,CIPLUS %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,CIPLUS %s
 
 ; GCN-LABEL: {{^}}local_i32_load
 ; SICIVI: s_mov_b32 m0
 ; GFX9-NOT: m0
 
 ; GCN: ds_read_b32 [[REG:v[0-9]+]], v{{[0-9]+}} offset:28
 ; GCN: buffer_store_dword [[REG]],
 define amdgpu_kernel void @local_i32_load(i32 addrspace(1)* %out, i32 addrspace(3)* %in) nounwind {
define amdgpu_kernel void @local_f64_store_0_offset(double addrspace(3)* %out) nounwind {
   ret void
 }
 
 ; GCN-LABEL: {{^}}local_v2i64_store:
 ; SICIVI: s_mov_b32 m0
 ; GFX9-NOT: m0
 
 ; GCN-NOT: add
-; GCN: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
+; SI: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:28 offset1:30
+; CIPLUS: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
 ; GCN: s_endpgm
 define amdgpu_kernel void @local_v2i64_store(<2 x i64> addrspace(3)* %out) nounwind {
   %gep = getelementptr <2 x i64>, <2 x i64> addrspace(3)* %out, i32 7
   store <2 x i64> <i64 5678, i64 5678>, <2 x i64> addrspace(3)* %gep, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}local_v2i64_store_0_offset:
 ; SICIVI: s_mov_b32 m0
 ; GFX9-NOT: m0
 
 ; GCN-NOT: add
-; GCN: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1
+; SI: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:2
+; CIPLUS: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1
 ; GCN: s_endpgm
 define amdgpu_kernel void @local_v2i64_store_0_offset(<2 x i64> addrspace(3)* %out) nounwind {
   store <2 x i64> <i64 1234, i64 1234>, <2 x i64> addrspace(3)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}local_v4i64_store:
 ; SICIVI: s_mov_b32 m0
 ; GFX9-NOT: m0
 
 ; GCN-NOT: add
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:30 offset1:31
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:28 offset1:29
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:60 offset1:62
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:56 offset1:58
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:30 offset1:31
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:28 offset1:29
 ; GCN: s_endpgm
 define amdgpu_kernel void @local_v4i64_store(<4 x i64> addrspace(3)* %out) nounwind {
   %gep = getelementptr <4 x i64>, <4 x i64> addrspace(3)* %out, i32 7
   store <4 x i64> <i64 5678, i64 5678, i64 5678, i64 5678>, <4 x i64> addrspace(3)* %gep, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}local_v4i64_store_0_offset:
 ; SICIVI: s_mov_b32 m0
 ; GFX9-NOT: m0
 
 ; GCN-NOT: add
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
-; GCN-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:4 offset1:6
+; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:2
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
+; CIPLUS-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1
 ; GCN: s_endpgm
 define amdgpu_kernel void @local_v4i64_store_0_offset(<4 x i64> addrspace(3)* %out) nounwind {
   store <4 x i64> <i64 1234, i64 1234, i64 1234, i64 1234>, <4 x i64> addrspace(3)* %out, align 16
   ret void
 }