AMDGPU/GFX10: Support DLC bit in llvm.amdgcn.s.buffer.load intrinsic

[nhaehnle]

Summary

Change-Id: Ie4c971462a7749740938c687144e77441dac2539

[nhaehnle]

Formatting

[rampitec]

LGTM, except check prefix.

test/CodeGen/AMDGPU/smrd.ll

3 ↗

(On Diff #201527)

This "GFX910" looks misleading. I suggest "GFX9_10".

llvm/trunk/include/llvm/IR/IntrinsicsAMDGPU.td

class AMDGPUBufferLoad : Intrinsic <
   AMDGPURsrcIntrinsic<0>;
 def int_amdgcn_buffer_load_format : AMDGPUBufferLoad;
 def int_amdgcn_buffer_load : AMDGPUBufferLoad;
 
 def int_amdgcn_s_buffer_load : Intrinsic <
   [llvm_any_ty],
   [llvm_v4i32_ty,     // rsrc(SGPR)
    llvm_i32_ty,       // byte offset(SGPR/imm)
-   llvm_i32_ty],      // cachepolicy(imm; bit 0 = glc)
+   llvm_i32_ty],      // cachepolicy(imm; bit 0 = glc, bit 2 = dlc)
   [IntrNoMem, ImmArg<2>]>,
   AMDGPURsrcIntrinsic<0>;
 
 class AMDGPUBufferStore : Intrinsic <
   [],
   [llvm_any_ty,       // vdata(VGPR)
    llvm_v4i32_ty,     // rsrc(SGPR)
    llvm_i32_ty,       // vindex(VGPR)

llvm/trunk/lib/Target/AMDGPU/SIISelLowering.h

private:
 
   SDValue LowerGlobalAddress(AMDGPUMachineFunction *MFI, SDValue Op,
                              SelectionDAG &DAG) const override;
   SDValue lowerImplicitZextParam(SelectionDAG &DAG, SDValue Op,
                                  MVT VT, unsigned Offset) const;
   SDValue lowerImage(SDValue Op, const AMDGPU::ImageDimIntrinsicInfo *Intr,
                      SelectionDAG &DAG) const;
   SDValue lowerSBuffer(EVT VT, SDLoc DL, SDValue Rsrc, SDValue Offset,
-                       SDValue GLC, SelectionDAG &DAG) const;
+                       SDValue GLC, SDValue DLC, SelectionDAG &DAG) const;
 
   SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_VOID(SDValue Op, SelectionDAG &DAG) const;
 
   // The raw.tbuffer and struct.tbuffer intrinsics have two offset args: offset
   // (the offset that is included in bounds checking and swizzling, to be split
   // between the instruction's voffset and immoffset fields) and soffset (the

llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp

return constructRetValue(DAG, NewNode,
                              DMaskLanes, NumVDataDwords, DL,
                              *DAG.getContext());
   }
 
   return SDValue(NewNode, 0);
 }
 
 SDValue SITargetLowering::lowerSBuffer(EVT VT, SDLoc DL, SDValue Rsrc,
-                                       SDValue Offset, SDValue GLC,
+                                       SDValue Offset, SDValue GLC, SDValue DLC,
                                        SelectionDAG &DAG) const {
   MachineFunction &MF = DAG.getMachineFunction();
   MachineMemOperand *MMO = MF.getMachineMemOperand(
       MachinePointerInfo(),
       MachineMemOperand::MOLoad | MachineMemOperand::MODereferenceable |
           MachineMemOperand::MOInvariant,
       VT.getStoreSize(), VT.getStoreSize());
 
   if (!Offset->isDivergent()) {
     SDValue Ops[] = {
         Rsrc,
         Offset, // Offset
-        GLC     // glc
+        GLC,
+        DLC,
     };
     return DAG.getMemIntrinsicNode(AMDGPUISD::SBUFFER_LOAD, DL,
                                    DAG.getVTList(VT), Ops, VT, MMO);
   }
 
   // We have a divergent offset. Emit a MUBUF buffer load instead. We can
   // assume that the buffer is unswizzled.
   SmallVector<SDValue, 4> Loads;
SDValue SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::r600_read_tidig_z:
     return loadInputValue(DAG, &AMDGPU::VGPR_32RegClass, MVT::i32,
                           SDLoc(DAG.getEntryNode()),
                           MFI->getArgInfo().WorkItemIDZ);
   case Intrinsic::amdgcn_wavefrontsize:
     return DAG.getConstant(MF.getSubtarget<GCNSubtarget>().getWavefrontSize(),
                            SDLoc(Op), MVT::i32);
   case Intrinsic::amdgcn_s_buffer_load: {
-    unsigned Cache = cast<ConstantSDNode>(Op.getOperand(3))->getZExtValue();
-    return lowerSBuffer(VT, DL, Op.getOperand(1), Op.getOperand(2),
-                        DAG.getTargetConstant(Cache & 1, DL, MVT::i1), DAG);
+    bool IsGFX10 = Subtarget->getGeneration() >= AMDGPUSubtarget::GFX10;
+    SDValue GLC;
+    SDValue DLC = DAG.getTargetConstant(0, DL, MVT::i1);
+    if (!parseCachePolicy(Op.getOperand(3), DAG, &GLC, nullptr,
+                          IsGFX10 ? &DLC : nullptr))
+      return Op;
+    return lowerSBuffer(VT, DL, Op.getOperand(1), Op.getOperand(2), GLC, DLC,
+                        DAG);
   }
   case Intrinsic::amdgcn_fdiv_fast:
     return lowerFDIV_FAST(Op, DAG);
   case Intrinsic::amdgcn_interp_mov: {
     SDValue M0 = copyToM0(DAG, DAG.getEntryNode(), DL, Op.getOperand(4));
     SDValue Glue = M0.getValue(1);
     return DAG.getNode(AMDGPUISD::INTERP_MOV, DL, MVT::f32, Op.getOperand(1),
                        Op.getOperand(2), Op.getOperand(3), Glue);

llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.td

 
 //===----------------------------------------------------------------------===//
 // SI DAG Nodes
 //===----------------------------------------------------------------------===//
 
 def AMDGPUclamp : SDNode<"AMDGPUISD::CLAMP", SDTFPUnaryOp>;
 
 def SIsbuffer_load : SDNode<"AMDGPUISD::SBUFFER_LOAD",
-  SDTypeProfile<1, 3, [SDTCisVT<1, v4i32>, SDTCisVT<2, i32>, SDTCisVT<3, i1>]>,
+  SDTypeProfile<1, 4, [SDTCisVT<1, v4i32>, SDTCisVT<2, i32>, SDTCisVT<3, i1>,
+                       SDTCisVT<4, i1>]>,
   [SDNPMayLoad, SDNPMemOperand]
 >;
 
 def SIds_ordered_count : SDNode<"AMDGPUISD::DS_ORDERED_COUNT",
   SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisVT<1, i32>, SDTCisVT<2, i16>]>,
   [SDNPMayLoad, SDNPMayStore, SDNPMemOperand, SDNPHasChain, SDNPInGlue]
 >;
 

llvm/trunk/lib/Target/AMDGPU/SMInstructions.td

def : GCNPat <
     (vt (smrd_load (i64 SReg_64:$sbase))),
     (vt (!cast<SM_Pseudo>(Instr#"_IMM") i64:$sbase, 0, 0, 0))
   >;
 }
 
 multiclass SMLoad_Pattern <string Instr, ValueType vt> {
   // 1. Offset as an immediate
   def : GCNPat <
-    (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm i32:$offset), i1:$glc),
-    (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, (as_i1imm $glc), 0))
+    (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm i32:$offset), i1:$glc, i1:$dlc),
+    (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, (as_i1imm $glc),
+                                        (as_i1imm $dlc)))
   >;
 
   // 2. 32-bit IMM offset on CI
   def : GCNPat <
-    (vt (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm32 i32:$offset), i1:$glc)),
-    (!cast<InstSI>(Instr#"_IMM_ci") $sbase, $offset, (as_i1imm $glc), 0)> {
+    (vt (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm32 i32:$offset), i1:$glc, i1:$dlc)),
+    (!cast<InstSI>(Instr#"_IMM_ci") $sbase, $offset, (as_i1imm $glc), (as_i1imm $dlc))> {
     let OtherPredicates = [isGFX7Only];
   }
 
   // 3. Offset loaded in an 32bit SGPR
   def : GCNPat <
-    (SIsbuffer_load v4i32:$sbase, i32:$offset, i1:$glc),
-    (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, (as_i1imm $glc), 0))
+    (SIsbuffer_load v4i32:$sbase, i32:$offset, i1:$glc, i1:$dlc),
+    (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, (as_i1imm $glc),
+                                         (as_i1imm $dlc)))
   >;
 }
 
 // Global and constant loads can be selected to either MUBUF or SMRD
 // instructions, but SMRD instructions are faster so we want the instruction
 // selector to prefer those.
 let AddedComplexity = 100 in {
 

llvm/trunk/test/CodeGen/AMDGPU/smrd-gfx10.ll

+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefixes=GCN %s
+
+; GCN-LABEL: {{^}}smrd_imm_dlc:
+; GCN: s_buffer_load_dword s0, s[0:3], 0x0 dlc ; encoding: [0x00,0x40,0x20,0xf4,0x00,0x00,0x00,0xfa]
+define amdgpu_ps float @smrd_imm_dlc(<4 x i32> inreg %desc) #0 {
+main_body:
+  %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 0, i32 4)
+  ret float %r
+}
+
+; GCN-LABEL: {{^}}smrd_sgpr_dlc:
+; GCN: s_buffer_load_dword s0, s[0:3], s4 dlc ; encoding: [0x00,0x40,0x20,0xf4,0x00,0x00,0x00,0x08]
+define amdgpu_ps float @smrd_sgpr_dlc(<4 x i32> inreg %desc, i32 inreg %offset) #0 {
+main_body:
+  %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 %offset, i32 4)
+  ret float %r
+}
+
+; GCN-LABEL: {{^}}smrd_imm_glc_dlc:
+; GCN: s_buffer_load_dword s0, s[0:3], 0x0 glc dlc ; encoding: [0x00,0x40,0x21,0xf4,0x00,0x00,0x00,0xfa]
+define amdgpu_ps float @smrd_imm_glc_dlc(<4 x i32> inreg %desc) #0 {
+main_body:
+  %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 0, i32 5)
+  ret float %r
+}
+
+; GCN-LABEL: {{^}}smrd_sgpr_glc_dlc:
+; GCN: s_buffer_load_dword s0, s[0:3], s4 glc dlc ; encoding: [0x00,0x40,0x21,0xf4,0x00,0x00,0x00,0x08]
+define amdgpu_ps float @smrd_sgpr_glc_dlc(<4 x i32> inreg %desc, i32 inreg %offset) #0 {
+main_body:
+  %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 %offset, i32 5)
+  ret float %r
+}
+
+declare float @llvm.amdgcn.s.buffer.load.f32(<4 x i32>, i32, i32)
+
+!0 = !{}

llvm/trunk/test/CodeGen/AMDGPU/smrd.ll

-; RUN: llc -march=amdgcn -mcpu=tahiti  -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=SI   -check-prefix=GCN -check-prefix=SICIVI -check-prefix=SICI -check-prefix=SIVIGFX9 %s
+; RUN: llc -march=amdgcn -mcpu=tahiti  -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=SI   -check-prefix=GCN -check-prefix=SICIVI -check-prefix=SICI -check-prefix=SIVIGFX9_10 %s
 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=CI   -check-prefix=GCN -check-prefix=SICIVI -check-prefix=SICI %s
-; RUN: llc -march=amdgcn -mcpu=tonga   -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=VI   -check-prefix=GCN -check-prefix=SICIVI -check-prefix=VIGFX9 -check-prefix=SIVIGFX9 %s
-; RUN: llc -march=amdgcn -mcpu=gfx900  -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=GFX9 -check-prefix=GCN -check-prefix=VIGFX9 -check-prefix=SIVIGFX9  %s
+; RUN: llc -march=amdgcn -mcpu=tonga   -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=VI   -check-prefix=GCN -check-prefix=SICIVI -check-prefix=VIGFX9_10 -check-prefix=SIVIGFX9_10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx900  -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=GFX9 -check-prefix=GCN -check-prefix=VIGFX9_10 -check-prefix=SIVIGFX9_10  %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefix=GFX10 -check-prefix=GCN -check-prefix=VIGFX9_10 -check-prefix=SIVIGFX9_10  %s
 
 ; SMRD load with an immediate offset.
 ; GCN-LABEL: {{^}}smrd0:
 ; SICI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x1 ; encoding: [0x01
-; VIGFX9: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x4
+; VIGFX9_10: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x4
 define amdgpu_kernel void @smrd0(i32 addrspace(1)* %out, i32 addrspace(4)* %ptr) #0 {
 entry:
   %tmp = getelementptr i32, i32 addrspace(4)* %ptr, i64 1
   %tmp1 = load i32, i32 addrspace(4)* %tmp
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void
 }
 
 ; SMRD load with the largest possible immediate offset.
 ; GCN-LABEL: {{^}}smrd1:
 ; SICI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xff ; encoding: [0xff,0x{{[0-9]+[137]}}
-; VIGFX9: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3fc
+; VIGFX9_10: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3fc
 define amdgpu_kernel void @smrd1(i32 addrspace(1)* %out, i32 addrspace(4)* %ptr) #0 {
 entry:
   %tmp = getelementptr i32, i32 addrspace(4)* %ptr, i64 255
   %tmp1 = load i32, i32 addrspace(4)* %tmp
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void
 }
 
 ; SMRD load with an offset greater than the largest possible immediate.
 ; GCN-LABEL: {{^}}smrd2:
 ; SI: s_movk_i32 s[[OFFSET:[0-9]]], 0x400
 ; SI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], s[[OFFSET]] ; encoding: [0x0[[OFFSET]]
 ; CI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x100
-; VIGFX9: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x400
+; VIGFX9_10: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x400
 ; GCN: s_endpgm
 define amdgpu_kernel void @smrd2(i32 addrspace(1)* %out, i32 addrspace(4)* %ptr) #0 {
 entry:
   %tmp = getelementptr i32, i32 addrspace(4)* %ptr, i64 256
   %tmp1 = load i32, i32 addrspace(4)* %tmp
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void
 }
entry:
   ret void
 }
 
 ; SMRD load with the largest possible immediate offset on VI
 ; GCN-LABEL: {{^}}smrd4:
 ; SI: s_mov_b32 [[OFFSET:s[0-9]+]], 0xffffc
 ; SI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
 ; CI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3ffff
-; VIGFX9: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xffffc
+; VIGFX9_10: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xffffc
 define amdgpu_kernel void @smrd4(i32 addrspace(1)* %out, i32 addrspace(4)* %ptr) #0 {
 entry:
   %tmp = getelementptr i32, i32 addrspace(4)* %ptr, i64 262143
   %tmp1 = load i32, i32 addrspace(4)* %tmp
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void
 }
 
 ; SMRD load with an offset greater than the largest possible immediate on VI
 ; GCN-LABEL: {{^}}smrd5:
-; SIVIGFX9: s_mov_b32 [[OFFSET:s[0-9]+]], 0x100000
-; SIVIGFX9: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
+; SIVIGFX9_10: s_mov_b32 [[OFFSET:s[0-9]+]], 0x100000
+; SIVIGFX9_10: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
 ; CI: s_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x40000
 ; GCN: s_endpgm
 define amdgpu_kernel void @smrd5(i32 addrspace(1)* %out, i32 addrspace(4)* %ptr) #0 {
 entry:
   %tmp = getelementptr i32, i32 addrspace(4)* %ptr, i64 262144
   %tmp1 = load i32, i32 addrspace(4)* %tmp
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void
main_body:
   %d3 = insertelement <4 x i32> %d2, i32 3, i32 3
   %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %d3, i32 0, i32 0)
   ret float %r
 }
 
 ; SMRD load using the load.const.v4i32 intrinsic with an immediate offset
 ; GCN-LABEL: {{^}}smrd_load_const0:
 ; SICI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x4 ; encoding: [0x04
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x10
+; VIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x10
 define amdgpu_ps void @smrd_load_const0(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 16, i32 0)
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %tmp21, i1 true, i1 true) #0
   ret void
 }
 
 ; SMRD load using the load.const.v4i32 intrinsic with the largest possible immediate
 ; offset.
 ; GCN-LABEL: {{^}}smrd_load_const1:
 ; SICI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xff ; encoding: [0xff
 ; SICI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xff glc ; encoding: [0xff
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3fc
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3fc glc
+; VIGFX9_10-DAG: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3fc ;
+; VIGFX9_10-DAG: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3fc glc ;
 define amdgpu_ps void @smrd_load_const1(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 1020, i32 0)
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp22, i32 1020, i32 1)
   %s.buffer.float = bitcast i32 %s.buffer to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %s.buffer.float, i1 true, i1 true) #0
   ret void
 }
 
 ; SMRD load using the load.const.v4i32 intrinsic with an offset greater than the
 ; largets possible immediate.
 ; immediate offset.
 ; GCN-LABEL: {{^}}smrd_load_const2:
 ; SI: s_movk_i32 s[[OFFSET:[0-9]]], 0x400
 ; SI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], s[[OFFSET]] ; encoding: [0x0[[OFFSET]]
 ; SI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], s[[OFFSET]] ; encoding: [0x0[[OFFSET]]
 ; CI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x100
 ; CI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x100
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x400
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x400
+; VIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x400
+; VIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x400
 define amdgpu_ps void @smrd_load_const2(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 1024, i32 0)
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp22, i32 1024, i32 0)
   %s.buffer.float = bitcast i32 %s.buffer to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %s.buffer.float, i1 true, i1 true) #0
   ret void
 }
 
 ; SMRD load with the largest possible immediate offset on VI
 ; GCN-LABEL: {{^}}smrd_load_const3:
 ; SI: s_mov_b32 [[OFFSET:s[0-9]+]], 0xffffc
 ; SI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
 ; SI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
 ; CI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3ffff
 ; CI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x3ffff
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xffffc
-; VIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xffffc
+; VIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xffffc
+; VIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0xffffc
 define amdgpu_ps void @smrd_load_const3(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 1048572, i32 0)
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp22, i32 1048572, i32 0)
   %s.buffer.float = bitcast i32 %s.buffer to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %s.buffer.float, i1 true, i1 true) #0
   ret void
 }
 
 ; SMRD load with an offset greater than the largest possible immediate on VI
 ; GCN-LABEL: {{^}}smrd_load_const4:
-; SIVIGFX9: s_mov_b32 [[OFFSET:s[0-9]+]], 0x100000
-; SIVIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
-; SIVIGFX9: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
+; SIVIGFX9_10: s_mov_b32 [[OFFSET:s[0-9]+]], 0x100000
+; SIVIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
+; SIVIGFX9_10: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], [[OFFSET]]
 ; CI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x40000
 ; CI: s_buffer_load_dword s{{[0-9]}}, s[{{[0-9]:[0-9]}}], 0x40000
 ; GCN: s_endpgm
 define amdgpu_ps void @smrd_load_const4(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 1048576, i32 0)
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp22, i32 1048576, i32 0)
   %s.buffer.float = bitcast i32 %s.buffer to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %s.buffer.float, i1 true, i1 true) #0
   ret void
 }
 
 ; dwordx2 s.buffer.load
 ; GCN-LABEL: {{^}}s_buffer_load_dwordx2:
-; VIGFX9: s_buffer_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
+; VIGFX9_10: s_buffer_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
 ; SICI: s_buffer_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x20
 define amdgpu_ps void @s_buffer_load_dwordx2(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call <2 x i32> @llvm.amdgcn.s.buffer.load.v2i32(<4 x i32> %tmp22, i32 128, i32 0)
   %s.buffer.0 = extractelement <2 x i32> %s.buffer, i32 0
   %s.buffer.0.float = bitcast i32 %s.buffer.0 to float
   %s.buffer.1 = extractelement <2 x i32> %s.buffer, i32 1
   %s.buffer.1.float = bitcast i32 %s.buffer.1 to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %s.buffer.0.float, float %s.buffer.1.float, float %s.buffer.0.float, float %s.buffer.1.float, i1 true, i1 true) #0
   ret void
 }
 
 ; dwordx4 s.buffer.load
 ; GCN-LABEL: {{^}}s_buffer_load_dwordx4:
-; VIGFX9: s_buffer_load_dwordx4 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
+; VIGFX9_10: s_buffer_load_dwordx4 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
 ; SICI: s_buffer_load_dwordx4 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x20
 define amdgpu_ps void @s_buffer_load_dwordx4(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call <4 x i32> @llvm.amdgcn.s.buffer.load.v4i32(<4 x i32> %tmp22, i32 128, i32 0)
   %s.buffer.0 = extractelement <4 x i32> %s.buffer, i32 0
   %s.buffer.0.float = bitcast i32 %s.buffer.0 to float
   %s.buffer.1 = extractelement <4 x i32> %s.buffer, i32 1
   %s.buffer.1.float = bitcast i32 %s.buffer.1 to float
   %s.buffer.2 = extractelement <4 x i32> %s.buffer, i32 2
   %s.buffer.2.float = bitcast i32 %s.buffer.2 to float
   %s.buffer.3 = extractelement <4 x i32> %s.buffer, i32 3
   %s.buffer.3.float = bitcast i32 %s.buffer.3 to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %s.buffer.0.float, float %s.buffer.1.float, float %s.buffer.2.float, float %s.buffer.3.float, i1 true, i1 true) #0
   ret void
 }
 
 ; dwordx8 s.buffer.load
 ; GCN-LABEL: {{^}}s_buffer_load_dwordx8:
-; VIGFX9: s_buffer_load_dwordx8 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
+; VIGFX9_10: s_buffer_load_dwordx8 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
 ; SICI: s_buffer_load_dwordx8 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x20
 define amdgpu_ps void @s_buffer_load_dwordx8(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call <8 x i32> @llvm.amdgcn.s.buffer.load.v8i32(<4 x i32> %tmp22, i32 128, i32 0)
   %s.buffer.0 = extractelement <8 x i32> %s.buffer, i32 0
   %s.buffer.0.float = bitcast i32 %s.buffer.0 to float
   %s.buffer.1 = extractelement <8 x i32> %s.buffer, i32 2
   %s.buffer.1.float = bitcast i32 %s.buffer.1 to float
   %s.buffer.2 = extractelement <8 x i32> %s.buffer, i32 5
   %s.buffer.2.float = bitcast i32 %s.buffer.2 to float
   %s.buffer.3 = extractelement <8 x i32> %s.buffer, i32 7
   %s.buffer.3.float = bitcast i32 %s.buffer.3 to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %s.buffer.0.float, float %s.buffer.1.float, float %s.buffer.2.float, float %s.buffer.3.float, i1 true, i1 true) #0
   ret void
 }
 
 ; dwordx16 s.buffer.load
 ; GCN-LABEL: {{^}}s_buffer_load_dwordx16:
-; VIGFX9: s_buffer_load_dwordx16 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
+; VIGFX9_10: s_buffer_load_dwordx16 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x80
 ; SICI: s_buffer_load_dwordx16 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]:[0-9]}}], 0x20
 define amdgpu_ps void @s_buffer_load_dwordx16(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in) #0 {
 main_body:
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call <16 x i32> @llvm.amdgcn.s.buffer.load.v16i32(<4 x i32> %tmp22, i32 128, i32 0)
   %s.buffer.0 = extractelement <16 x i32> %s.buffer, i32 0
   %s.buffer.0.float = bitcast i32 %s.buffer.0 to float
   %s.buffer.1 = extractelement <16 x i32> %s.buffer, i32 3
main_body:
   %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 %off, i32 0)
   ret float %r
 }
 
 ; GCN-LABEL: {{^}}smrd_vgpr_offset_imm_too_large:
 ; GCN-NEXT: %bb.
 ; SICI-NEXT: v_add_{{i|u}}32_e32 v0, {{(vcc, )?}}0x1000, v0
 ; SICI-NEXT: buffer_load_dword v{{[0-9]}}, v0, s[0:3], 0 offen ;
-; VIGFX9-NEXT: buffer_load_dword v{{[0-9]}}, v0, s[0:3], 4 offen offset:4092 ;
+; VIGFX9_10-NEXT: buffer_load_dword v{{[0-9]}}, v0, s[0:3], 4 offen offset:4092 ;
 define amdgpu_ps float @smrd_vgpr_offset_imm_too_large(<4 x i32> inreg %desc, i32 %offset) #0 {
 main_body:
   %off = add i32 %offset, 4096
   %r = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 %off, i32 0)
   ret float %r
 }
 
 ; GCN-LABEL: {{^}}smrd_imm_merged:
 ; GCN-NEXT: %bb.
 ; SICI-NEXT: s_buffer_load_dwordx4 s[{{[0-9]}}:{{[0-9]}}], s[0:3], 0x1
 ; SICI-NEXT: s_buffer_load_dwordx2 s[{{[0-9]}}:{{[0-9]}}], s[0:3], 0x7
-; VIGFX9-NEXT: s_buffer_load_dwordx4 s[{{[0-9]}}:{{[0-9]}}], s[0:3], 0x4
-; VIGFX9-NEXT: s_buffer_load_dwordx2 s[{{[0-9]}}:{{[0-9]}}], s[0:3], 0x1c
+; VIGFX9_10-NEXT: s_buffer_load_dwordx4 s[{{[0-9]}}:{{[0-9]}}], s[0:3], 0x4
+; VIGFX9_10-NEXT: s_buffer_load_dwordx2 s[{{[0-9]}}:{{[0-9]}}], s[0:3], 0x1c
 define amdgpu_ps void @smrd_imm_merged(<4 x i32> inreg %desc) #0 {
 main_body:
   %r1 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 4, i32 0)
   %r2 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 8, i32 0)
   %r3 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 12, i32 0)
   %r4 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 16, i32 0)
   %r5 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 28, i32 0)
   %r6 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %desc, i32 32, i32 0)
.outer_loop_body:
   %offset = shl i32 %loopctr.2, 6
   %load2result = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %descriptor, i32 %offset, i32 0)
   %outer_br = fcmp ueq float %load2result, 0x0
   br i1 %outer_br, label %.outer_loop_header, label %ret_block
 }
 
 ; SMRD load with a non-const offset
 ; GCN-LABEL: {{^}}smrd_load_nonconst0:
-; SIVIGFX9: s_buffer_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
-; SIVIGFX9: s_buffer_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
+; SIVIGFX9_10: s_buffer_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
+; SIVIGFX9_10: s_buffer_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
 ; CI: s_buffer_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
 ; CI: s_buffer_load_dword s{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
 ; GCN: s_endpgm
 define amdgpu_ps void @smrd_load_nonconst0(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in, i32 inreg %ncoff) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 %ncoff, i32 0)
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp22, i32 %ncoff, i32 0)
   %s.buffer.float = bitcast i32 %s.buffer to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %s.buffer.float, i1 true, i1 true) #0
   ret void
 }
 
 ; SMRD load with a non-const non-uniform offset
 ; GCN-LABEL: {{^}}smrd_load_nonconst1:
-; SIVIGFX9: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
-; SIVIGFX9: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
+; SIVIGFX9_10: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
+; SIVIGFX9_10: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
 ; CI: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
 ; CI: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
 ; GCN: s_endpgm
 define amdgpu_ps void @smrd_load_nonconst1(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in, i32 %ncoff) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 %ncoff, i32 0)
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %tmp22, i32 %ncoff, i32 0)
   %s.buffer.float = bitcast i32 %s.buffer to float
   call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp21, float %tmp21, float %tmp21, float %s.buffer.float, i1 true, i1 true) #0
   ret void
 }
 
 ; SMRD load with a non-const non-uniform offset of > 4 dwords (requires splitting)
 ; GCN-LABEL: {{^}}smrd_load_nonconst2:
-; SIVIGFX9: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
-; SIVIGFX9: buffer_load_dwordx4 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
+; SIVIGFX9_10: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
+; SIVIGFX9_10: buffer_load_dwordx4 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
 ; CI: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
 ; CI: buffer_load_dwordx4 v[{{[0-9]+:[0-9]+}}], v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], 0 offen
 ; GCN: s_endpgm
 define amdgpu_ps void @smrd_load_nonconst2(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in, i32 %ncoff) #0 {
 main_body:
   %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(4)* %arg, i32 0
   %tmp20 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp
   %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 %ncoff, i32 0)

 }
 
 ; GCN-LABEL: {{^}}smrd_load_nonconst4:
 ; SICI: v_add_i32_e32 v{{[0-9]+}}, vcc, 0xff8, v0 ;
 ; SICI-DAG: buffer_load_dwordx4 v[0:3], v{{[0-9]+}}, s[0:3], 0 offen ;
 ; SICI-DAG: buffer_load_dwordx4 v[4:7], v{{[0-9]+}}, s[0:3], 0 offen offset:16 ;
 ; SICI-DAG: buffer_load_dwordx4 v[8:11], v{{[0-9]+}}, s[0:3], 0 offen offset:32 ;
 ; SICI-DAG: buffer_load_dwordx4 v[12:15], v{{[0-9]+}}, s[0:3], 0 offen offset:48 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[0:3], v{{[0-9]+}}, s[0:3], 56 offen offset:4032 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[4:7], v{{[0-9]+}}, s[0:3], 56 offen offset:4048 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[8:11], v{{[0-9]+}}, s[0:3], 56 offen offset:4064 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[12:15], v{{[0-9]+}}, s[0:3], 56 offen offset:4080 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[0:3], v{{[0-9]+}}, s[0:3], 56 offen offset:4032 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[4:7], v{{[0-9]+}}, s[0:3], 56 offen offset:4048 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[8:11], v{{[0-9]+}}, s[0:3], 56 offen offset:4064 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[12:15], v{{[0-9]+}}, s[0:3], 56 offen offset:4080 ;
 ; GCN: ; return to shader part epilog
 define amdgpu_ps <16 x float> @smrd_load_nonconst4(<4 x i32> inreg %rsrc, i32 %off) #0 {
 main_body:
   %off.2 = add i32 %off, 4088
   %ld = call <16 x i32> @llvm.amdgcn.s.buffer.load.v16i32(<4 x i32> %rsrc, i32 %off.2, i32 0)
   %bc = bitcast <16 x i32> %ld to <16 x float>
   ret <16 x float> %bc
 }
 
 ; GCN-LABEL: {{^}}smrd_load_nonconst5:
 ; SICI: v_add_i32_e32 v{{[0-9]+}}, vcc, 0x1004, v0
 ; SICI-DAG: buffer_load_dwordx4 v[0:3], v{{[0-9]+}}, s[0:3], 0 offen ;
 ; SICI-DAG: buffer_load_dwordx4 v[4:7], v{{[0-9]+}}, s[0:3], 0 offen offset:16 ;
 ; SICI-DAG: buffer_load_dwordx4 v[8:11], v{{[0-9]+}}, s[0:3], 0 offen offset:32 ;
 ; SICI-DAG: buffer_load_dwordx4 v[12:15], v{{[0-9]+}}, s[0:3], 0 offen offset:48 ;
-; VIGFX9: s_movk_i32 s4, 0xfc0
-; VIGFX9-DAG: buffer_load_dwordx4 v[0:3], v{{[0-9]+}}, s[0:3], s4 offen offset:68 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[4:7], v{{[0-9]+}}, s[0:3], s4 offen offset:84 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[8:11], v{{[0-9]+}}, s[0:3], s4 offen offset:100 ;
-; VIGFX9-DAG: buffer_load_dwordx4 v[12:15], v{{[0-9]+}}, s[0:3], s4 offen offset:116 ;
+; VIGFX9_10: s_movk_i32 s4, 0xfc0
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[0:3], v{{[0-9]+}}, s[0:3], s4 offen offset:68 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[4:7], v{{[0-9]+}}, s[0:3], s4 offen offset:84 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[8:11], v{{[0-9]+}}, s[0:3], s4 offen offset:100 ;
+; VIGFX9_10-DAG: buffer_load_dwordx4 v[12:15], v{{[0-9]+}}, s[0:3], s4 offen offset:116 ;
 ; GCN: ; return to shader part epilog
 define amdgpu_ps <16 x float> @smrd_load_nonconst5(<4 x i32> inreg %rsrc, i32 %off) #0 {
 main_body:
   %off.2 = add i32 %off, 4100
   %ld = call <16 x i32> @llvm.amdgcn.s.buffer.load.v16i32(<4 x i32> %rsrc, i32 %off.2, i32 0)
   %bc = bitcast <16 x i32> %ld to <16 x float>
   ret <16 x float> %bc
 }
 
 ; SMRD load dwordx2
 ; GCN-LABEL: {{^}}smrd_load_dwordx2:
-; SIVIGFX9: s_buffer_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
+; SIVIGFX9_10: s_buffer_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
 ; CI: s_buffer_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
 ; GCN: s_endpgm
 define amdgpu_ps void @smrd_load_dwordx2(<4 x i32> addrspace(4)* inreg %arg, <4 x i32> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, <4 x i32> addrspace(4)* inreg %in, i32 inreg %ncoff) #0 {
 main_body:
   %tmp22 = load <4 x i32>, <4 x i32> addrspace(4)* %in
   %s.buffer = call <2 x i32> @llvm.amdgcn.s.buffer.load.v2i32(<4 x i32> %tmp22, i32 %ncoff, i32 0)
   %s.buffer.float = bitcast <2 x i32> %s.buffer to <2 x float>
   %r.1 = extractelement <2 x float> %s.buffer.float, i32 0