AMDGPU: Start adding offset fields to flat instructions

llvm-svn: 305194

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

@@ -138,7 +138,8 @@ private:
   bool SelectMUBUFIntrinsicVOffset(SDValue Offset, SDValue &SOffset,
                                    SDValue &ImmOffset, SDValue &VOffset) const;
 
-  bool SelectFlat(SDValue Addr, SDValue &VAddr, SDValue &SLC) const;
+  bool SelectFlat(SDValue Addr, SDValue &VAddr,
+                  SDValue &Offset, SDValue &SLC) const;
 
   bool SelectSMRDOffset(SDValue ByteOffsetNode, SDValue &Offset,
                         bool &Imm) const;
@@ -1315,8 +1316,10 @@ bool AMDGPUDAGToDAGISel::SelectMUBUFIntrinsicVOffset(SDValue Offset,
 
 bool AMDGPUDAGToDAGISel::SelectFlat(SDValue Addr,
                                     SDValue &VAddr,
+                                    SDValue &Offset,
                                     SDValue &SLC) const {
   VAddr = Addr;
+  Offset = CurDAG->getTargetConstant(0, SDLoc(), MVT::i16);
   SLC = CurDAG->getTargetConstant(0, SDLoc(), MVT::i1);
   return true;
 }

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

@@ -126,8 +126,9 @@ bool AMDGPUInstructionSelector::selectG_STORE(MachineInstr &I) const {
   MachineInstr *Flat = BuildMI(*BB, &I, DL, TII.get(AMDGPU::FLAT_STORE_DWORD))
           .add(I.getOperand(1))
           .add(I.getOperand(0))
-          .addImm(0)
+          .addImm(0)  // offset
-          .addImm(0);
+          .addImm(0)  // glc
+          .addImm(0); // slc
 
 
   // Now that we selected an opcode, we need to constrain the register
@@ -392,8 +393,9 @@ bool AMDGPUInstructionSelector::selectG_LOAD(MachineInstr &I) const {
   MachineInstr *Flat = BuildMI(*BB, &I, DL, TII.get(Opcode))
                                .add(I.getOperand(0))
                                .addReg(PtrReg)
-                               .addImm(0)
+                               .addImm(0)  // offset
-                               .addImm(0);
+                               .addImm(0)  // glc
+                               .addImm(0); // slc
 
   bool Ret = constrainSelectedInstRegOperands(*Flat, TII, TRI, RBI);
   I.eraseFromParent();

llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

@@ -285,6 +285,9 @@ public:
   bool isOffset() const { return isImmTy(ImmTyOffset) && isUInt<16>(getImm()); }
   bool isOffset0() const { return isImmTy(ImmTyOffset0) && isUInt<16>(getImm()); }
   bool isOffset1() const { return isImmTy(ImmTyOffset1) && isUInt<8>(getImm()); }
+
+  bool isOffsetU12() const { return isImmTy(ImmTyOffset) && isUInt<12>(getImm()); }
+  bool isOffsetS13() const { return isImmTy(ImmTyOffset) && isInt<13>(getImm()); }
   bool isGDS() const { return isImmTy(ImmTyGDS); }
   bool isGLC() const { return isImmTy(ImmTyGLC); }
   bool isSLC() const { return isImmTy(ImmTySLC); }
@@ -886,6 +889,10 @@ public:
     return getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm];
   }
 
+  bool hasFlatOffsets() const {
+    return getFeatureBits()[AMDGPU::FeatureFlatInstOffsets];
+  }
+
   bool hasSGPR102_SGPR103() const {
     return !isVI();
   }
@@ -1034,6 +1041,8 @@ public:
   AMDGPUOperand::Ptr defaultSMRDOffset8() const;
   AMDGPUOperand::Ptr defaultSMRDOffset20() const;
   AMDGPUOperand::Ptr defaultSMRDLiteralOffset() const;
+  AMDGPUOperand::Ptr defaultOffsetU12() const;
+  AMDGPUOperand::Ptr defaultOffsetS13() const;
 
   OperandMatchResultTy parseOModOperand(OperandVector &Operands);
 
@@ -1970,6 +1979,15 @@ unsigned AMDGPUAsmParser::checkTargetMatchPredicate(MCInst &Inst) {
     }
   }
 
+  if ((TSFlags & SIInstrFlags::FLAT) && !hasFlatOffsets()) {
+    // FIXME: Produces error without correct column reported.
+    auto OpNum =
+        AMDGPU::getNamedOperandIdx(Inst.getOpcode(), AMDGPU::OpName::offset);
+    const auto &Op = Inst.getOperand(OpNum);
+    if (Op.getImm() != 0)
+      return Match_InvalidOperand;
+  }
+
   return Match_Success;
 }
 
@@ -3849,6 +3867,14 @@ AMDGPUOperand::Ptr AMDGPUAsmParser::defaultSMRDLiteralOffset() const {
   return AMDGPUOperand::CreateImm(this, 0, SMLoc(), AMDGPUOperand::ImmTyOffset);
 }
 
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultOffsetU12() const {
+  return AMDGPUOperand::CreateImm(this, 0, SMLoc(), AMDGPUOperand::ImmTyOffset);
+}
+
+AMDGPUOperand::Ptr AMDGPUAsmParser::defaultOffsetS13() const {
+  return AMDGPUOperand::CreateImm(this, 0, SMLoc(), AMDGPUOperand::ImmTyOffset);
+}
+
 //===----------------------------------------------------------------------===//
 // vop3
 //===----------------------------------------------------------------------===//

llvm/lib/Target/AMDGPU/FLATInstructions.td

@@ -7,7 +7,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-def FLATAtomic : ComplexPattern<i64, 2, "SelectFlat">;
+def FLATAtomic : ComplexPattern<i64, 3, "SelectFlat">;
 
 //===----------------------------------------------------------------------===//
 // FLAT classes
@@ -55,6 +55,8 @@ class FLAT_Real <bits<7> op, FLAT_Pseudo ps> :
   // copy relevant pseudo op flags
   let SubtargetPredicate = ps.SubtargetPredicate;
   let AsmMatchConverter  = ps.AsmMatchConverter;
+  let TSFlags = ps.TSFlags;
+  let UseNamedOperandTable = ps.UseNamedOperandTable;
 
   // encoding fields
   bits<8> vaddr;
@@ -63,10 +65,23 @@ class FLAT_Real <bits<7> op, FLAT_Pseudo ps> :
   bits<1> slc;
   bits<1> glc;
 
+  // Only valid on gfx9
+  bits<1> lds = 0; // XXX - What does this actually do?
+  bits<2> seg; // Segment, 00=flat, 01=scratch, 10=global, 11=reserved
+
+  // Signed offset. Highest bit ignored for flat and treated as 12-bit
+  // unsigned for flat acceses.
+  bits<13> offset;
+  bits<1> nv = 0; // XXX - What does this actually do?
+
   // We don't use tfe right now, and it was removed in gfx9.
   bits<1> tfe = 0;
 
-  // 15-0 is reserved.
+  // Only valid on GFX9+
+  let Inst{12-0} = offset;
+  let Inst{13} = lds;
+  let Inst{15-14} = 0;
+
   let Inst{16}    = !if(ps.has_glc, glc, ps.glcValue);
   let Inst{17}    = slc;
   let Inst{24-18} = op;
@@ -74,24 +89,30 @@ class FLAT_Real <bits<7> op, FLAT_Pseudo ps> :
   let Inst{39-32} = vaddr;
   let Inst{47-40} = !if(ps.has_data, vdata, ?);
   // 54-48 is reserved.
-  let Inst{55}    = tfe;
+  let Inst{55}    = nv; // nv on GFX9+, TFE before.
   let Inst{63-56} = !if(ps.has_vdst, vdst, ?);
 }
 
-class FLAT_Load_Pseudo <string opName, RegisterClass regClass> : FLAT_Pseudo<
+class FLAT_Load_Pseudo <string opName, RegisterClass regClass,
+  bit HasSignedOffset = 0> : FLAT_Pseudo<
   opName,
   (outs regClass:$vdst),
-  (ins VReg_64:$vaddr, GLC:$glc, slc:$slc),
+  !if(HasSignedOffset,
-  " $vdst, $vaddr$glc$slc"> {
+    (ins VReg_64:$vaddr, offset_s13:$offset, GLC:$glc, slc:$slc),
+    (ins VReg_64:$vaddr, offset_u12:$offset, GLC:$glc, slc:$slc)),
+  " $vdst, $vaddr$offset$glc$slc"> {
   let has_data = 0;
   let mayLoad = 1;
 }
 
-class FLAT_Store_Pseudo <string opName, RegisterClass vdataClass> : FLAT_Pseudo<
+class FLAT_Store_Pseudo <string opName, RegisterClass vdataClass,
+  bit HasSignedOffset = 0> : FLAT_Pseudo<
   opName,
   (outs),
-  (ins VReg_64:$vaddr, vdataClass:$vdata, GLC:$glc, slc:$slc),
+  !if(HasSignedOffset,
-  " $vaddr, $vdata$glc$slc"> {
+    (ins VReg_64:$vaddr, vdataClass:$vdata, offset_s13:$offset, GLC:$glc, slc:$slc),
+    (ins VReg_64:$vaddr, vdataClass:$vdata, offset_u12:$offset, GLC:$glc, slc:$slc)),
+  " $vaddr, $vdata$offset$glc$slc"> {
   let mayLoad  = 0;
   let mayStore = 1;
   let has_vdst = 0;
@@ -103,12 +124,15 @@ multiclass FLAT_Atomic_Pseudo<
   ValueType vt,
   SDPatternOperator atomic = null_frag,
   ValueType data_vt = vt,
-  RegisterClass data_rc = vdst_rc> {
+  RegisterClass data_rc = vdst_rc,
+  bit HasSignedOffset = 0> {
 
   def "" : FLAT_Pseudo <opName,
     (outs),
-    (ins VReg_64:$vaddr, data_rc:$vdata, slc:$slc),
+    !if(HasSignedOffset,
-    " $vaddr, $vdata$slc",
+      (ins VReg_64:$vaddr, data_rc:$vdata, offset_s13:$offset, slc:$slc),
+      (ins VReg_64:$vaddr, data_rc:$vdata, offset_u12:$offset, slc:$slc)),
+    " $vaddr, $vdata$offset$slc",
     []>,
     AtomicNoRet <NAME, 0> {
     let mayLoad = 1;
@@ -121,10 +145,12 @@ multiclass FLAT_Atomic_Pseudo<
 
   def _RTN : FLAT_Pseudo <opName,
     (outs vdst_rc:$vdst),
-    (ins VReg_64:$vaddr, data_rc:$vdata, slc:$slc),
+    !if(HasSignedOffset,
-    " $vdst, $vaddr, $vdata glc$slc",
+      (ins VReg_64:$vaddr, data_rc:$vdata, offset_s13:$offset, slc:$slc),
+      (ins VReg_64:$vaddr, data_rc:$vdata, offset_u12:$offset, slc:$slc)),
+    " $vdst, $vaddr, $vdata$offset glc$slc",
     [(set vt:$vdst,
-      (atomic (FLATAtomic i64:$vaddr, i1:$slc), data_vt:$vdata))]>,
+      (atomic (FLATAtomic i64:$vaddr, i16:$offset, i1:$slc), data_vt:$vdata))]>,
     AtomicNoRet <NAME, 1> {
     let mayLoad  = 1;
     let mayStore = 1;
@@ -313,30 +339,30 @@ def flat_truncstorei16 : flat_st <truncstorei16>;
 // Patterns for global loads with no offset.
 class FlatLoadPat <FLAT_Pseudo inst, SDPatternOperator node, ValueType vt> : Pat <
   (vt (node i64:$addr)),
-  (inst $addr, 0, 0)
+  (inst $addr, 0, 0, 0)
 >;
 
 class FlatLoadAtomicPat <FLAT_Pseudo inst, SDPatternOperator node, ValueType vt> : Pat <
   (vt (node i64:$addr)),
-  (inst $addr, 1, 0)
+  (inst $addr, 0, 1, 0)
 >;
 
 class FlatStorePat <FLAT_Pseudo inst, SDPatternOperator node, ValueType vt> : Pat <
   (node vt:$data, i64:$addr),
-  (inst $addr, $data, 0, 0)
+  (inst $addr, $data, 0, 0, 0)
 >;
 
 class FlatStoreAtomicPat <FLAT_Pseudo inst, SDPatternOperator node, ValueType vt> : Pat <
   // atomic store follows atomic binop convention so the address comes
   // first.
   (node i64:$addr, vt:$data),
-  (inst $addr, $data, 1, 0)
+  (inst $addr, $data, 0, 1, 0)
 >;
 
 class FlatAtomicPat <FLAT_Pseudo inst, SDPatternOperator node, ValueType vt,
                      ValueType data_vt = vt> : Pat <
   (vt (node i64:$addr, data_vt:$data)),
-  (inst $addr, $data, 0)
+  (inst $addr, $data, 0, 0)
 >;
 
 let Predicates = [isCIVI] in {

llvm/lib/Target/AMDGPU/SIInstrInfo.td

@@ -492,11 +492,21 @@ class NamedOperandU8<string Name, AsmOperandClass MatchClass> : Operand<i8> {
   let ParserMatchClass = MatchClass;
 }
 
+class NamedOperandU12<string Name, AsmOperandClass MatchClass> : Operand<i16> {
+  let PrintMethod = "print"#Name;
+  let ParserMatchClass = MatchClass;
+}
+
 class NamedOperandU16<string Name, AsmOperandClass MatchClass> : Operand<i16> {
   let PrintMethod = "print"#Name;
   let ParserMatchClass = MatchClass;
 }
 
+class NamedOperandS13<string Name, AsmOperandClass MatchClass> : Operand<i16> {
+  let PrintMethod = "print"#Name;
+  let ParserMatchClass = MatchClass;
+}
+
 class NamedOperandU32<string Name, AsmOperandClass MatchClass> : Operand<i32> {
   let PrintMethod = "print"#Name;
   let ParserMatchClass = MatchClass;
@@ -514,6 +524,8 @@ def offen : NamedOperandBit<"Offen", NamedMatchClass<"Offen">>;
 def idxen : NamedOperandBit<"Idxen", NamedMatchClass<"Idxen">>;
 def addr64 : NamedOperandBit<"Addr64", NamedMatchClass<"Addr64">>;
 
+def offset_u12 : NamedOperandU12<"Offset", NamedMatchClass<"OffsetU12">>;
+def offset_s13 : NamedOperandS13<"Offset", NamedMatchClass<"OffsetS13">>;
 def offset : NamedOperandU16<"Offset", NamedMatchClass<"Offset">>;
 def offset0 : NamedOperandU8<"Offset0", NamedMatchClass<"Offset0">>;
 def offset1 : NamedOperandU8<"Offset1", NamedMatchClass<"Offset1">>;

llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir

@@ -14,7 +14,7 @@ regBankSelected: true
 
 # GCN: global_addrspace
 # GCN: [[PTR:%[0-9]+]] = COPY %vgpr0_vgpr1
-# GCN: FLAT_LOAD_DWORD  [[PTR]], 0, 0
+# GCN: FLAT_LOAD_DWORD  [[PTR]], 0, 0, 0
 
 body: |
   bb.0:

llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-store-flat.mir

@@ -15,7 +15,7 @@ regBankSelected: true
 # GCN: global_addrspace
 # GCN: [[PTR:%[0-9]+]] = COPY %vgpr0_vgpr1
 # GCN: [[VAL:%[0-9]+]] = COPY %vgpr2
-# GCN: FLAT_STORE_DWORD [[PTR]], [[VAL]], 0, 0
+# GCN: FLAT_STORE_DWORD [[PTR]], [[VAL]], 0, 0, 0
 
 body: |
   bb.0:

llvm/test/CodeGen/AMDGPU/constant-fold-imm-immreg.mir

@@ -219,19 +219,19 @@ body:             |
     %34 = V_MOV_B32_e32 63, implicit %exec
 
     %27 = V_AND_B32_e64 %26, %24, implicit %exec
-    FLAT_STORE_DWORD %37, %27, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %37, %27, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %28 = V_AND_B32_e64 %24, %26, implicit %exec
-    FLAT_STORE_DWORD %37, %28, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %37, %28, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %29 = V_AND_B32_e32 %26, %24, implicit %exec
-    FLAT_STORE_DWORD %37, %29, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %37, %29, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %30 = V_AND_B32_e64 %26, %26, implicit %exec
-    FLAT_STORE_DWORD %37, %30, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %37, %30, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %31 = V_AND_B32_e64 %34, %34, implicit %exec
-    FLAT_STORE_DWORD %37, %31, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %37, %31, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     S_ENDPGM
 
@@ -407,34 +407,34 @@ body:             |
     %27 = S_MOV_B32 -4
 
     %11 = V_LSHLREV_B32_e64 12, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %11, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %11, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %12 = V_LSHLREV_B32_e64 %7, 12, implicit %exec
-    FLAT_STORE_DWORD %20, %12, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %12, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %13 = V_LSHL_B32_e64 %7, 12, implicit %exec
-    FLAT_STORE_DWORD %20, %13, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %13, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %14 = V_LSHL_B32_e64 12, %7, implicit %exec
-    FLAT_STORE_DWORD %20, %14, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %14, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %15 = V_LSHL_B32_e64 12, %24, implicit %exec
-    FLAT_STORE_DWORD %20, %15, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %15, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %22 = V_LSHL_B32_e64 %6, 12, implicit %exec
-    FLAT_STORE_DWORD %20, %22, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %22, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %23 = V_LSHL_B32_e64 %6, 32, implicit %exec
-    FLAT_STORE_DWORD %20, %23, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %23, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %25 = V_LSHL_B32_e32 %6, %6, implicit %exec
-    FLAT_STORE_DWORD %20, %25, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %25, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %26 = V_LSHLREV_B32_e32 11, %24, implicit %exec
-    FLAT_STORE_DWORD %20, %26, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %26, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %28 = V_LSHL_B32_e32 %27, %6, implicit %exec
-    FLAT_STORE_DWORD %20, %28, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %28, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     S_ENDPGM
 
@@ -615,34 +615,34 @@ body:             |
     %35 = V_MOV_B32_e32 2, implicit %exec
 
     %11 = V_ASHRREV_I32_e64 8, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %11, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %11, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %12 = V_ASHRREV_I32_e64 %8, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %12, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %12, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %13 = V_ASHR_I32_e64 %7, 3, implicit %exec
-    FLAT_STORE_DWORD %20, %13, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %13, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %14 = V_ASHR_I32_e64 7, %32, implicit %exec
-    FLAT_STORE_DWORD %20, %14, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %14, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %15 = V_ASHR_I32_e64 %27, %24, implicit %exec
-    FLAT_STORE_DWORD %20, %15, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %15, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %22 = V_ASHR_I32_e64 %6, 4, implicit %exec
-    FLAT_STORE_DWORD %20, %22, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %22, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %23 = V_ASHR_I32_e64 %6, %33, implicit %exec
-    FLAT_STORE_DWORD %20, %23, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %23, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %25 = V_ASHR_I32_e32 %34, %34, implicit %exec
-    FLAT_STORE_DWORD %20, %25, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %25, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %26 = V_ASHRREV_I32_e32 11, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %26, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %26, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %28 = V_ASHR_I32_e32 %27, %35, implicit %exec
-    FLAT_STORE_DWORD %20, %28, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %28, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     S_ENDPGM
 
@@ -824,34 +824,34 @@ body:             |
     %35 = V_MOV_B32_e32 2, implicit %exec
 
     %11 = V_LSHRREV_B32_e64 8, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %11, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %11, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %12 = V_LSHRREV_B32_e64 %8, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %12, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %12, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %13 = V_LSHR_B32_e64 %7, 3, implicit %exec
-    FLAT_STORE_DWORD %20, %13, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %13, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %14 = V_LSHR_B32_e64 7, %32, implicit %exec
-    FLAT_STORE_DWORD %20, %14, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %14, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %15 = V_LSHR_B32_e64 %27, %24, implicit %exec
-    FLAT_STORE_DWORD %20, %15, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %15, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %22 = V_LSHR_B32_e64 %6, 4, implicit %exec
-    FLAT_STORE_DWORD %20, %22, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %22, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %23 = V_LSHR_B32_e64 %6, %33, implicit %exec
-    FLAT_STORE_DWORD %20, %23, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %23, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %25 = V_LSHR_B32_e32 %34, %34, implicit %exec
-    FLAT_STORE_DWORD %20, %25, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %25, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %26 = V_LSHRREV_B32_e32 11, %10, implicit %exec
-    FLAT_STORE_DWORD %20, %26, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %26, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     %28 = V_LSHR_B32_e32 %27, %35, implicit %exec
-    FLAT_STORE_DWORD %20, %28, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
+    FLAT_STORE_DWORD %20, %28, 0, 0, 0, implicit %exec, implicit %flat_scr :: (volatile store 4 into %ir.gep.out)
 
     S_ENDPGM
 

llvm/test/CodeGen/AMDGPU/inserted-wait-states.mir

@@ -246,15 +246,15 @@ body: |
     S_BRANCH %bb.1
 
   bb.1:
-    FLAT_STORE_DWORDX2 %vgpr0_vgpr1, %vgpr2_vgpr3, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX2 %vgpr0_vgpr1, %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
-    FLAT_STORE_DWORDX3 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX3 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
-    FLAT_STORE_DWORDX4 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX4 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
-    FLAT_ATOMIC_CMPSWAP_X2 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, implicit %exec, implicit %flat_scr
+    FLAT_ATOMIC_CMPSWAP_X2 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
-    FLAT_ATOMIC_FCMPSWAP_X2 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, implicit %exec, implicit %flat_scr
+    FLAT_ATOMIC_FCMPSWAP_X2 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
     S_ENDPGM
 

llvm/test/CodeGen/AMDGPU/limit-coalesce.mir

@@ -57,15 +57,15 @@ body:             |
     %4.sub1 = COPY %3.sub0
     undef %5.sub0 = COPY %4.sub1
     %5.sub1 = COPY %4.sub0
-    FLAT_STORE_DWORDX2 %vgpr0_vgpr1, killed %5, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX2 %vgpr0_vgpr1, killed %5, 0, 0, 0, implicit %exec, implicit %flat_scr
 
     %6 = IMPLICIT_DEF
     undef %7.sub0_sub1 = COPY %6
     %7.sub2 = COPY %3.sub0
-    FLAT_STORE_DWORDX3 %vgpr0_vgpr1, killed %7, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX3 %vgpr0_vgpr1, killed %7, 0, 0, 0, implicit %exec, implicit %flat_scr
 
     %8 = IMPLICIT_DEF
     undef %9.sub0_sub1_sub2 = COPY %8
     %9.sub3 = COPY %3.sub0
-    FLAT_STORE_DWORDX4 %vgpr0_vgpr1, killed %9, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX4 %vgpr0_vgpr1, killed %9, 0, 0, 0, implicit %exec, implicit %flat_scr
 ...

llvm/test/CodeGen/AMDGPU/rename-independent-subregs-invalid-mac-operands.mir

@@ -58,12 +58,12 @@ body:             |
 
   bb.3:
     %1 = COPY killed %17
-    FLAT_STORE_DWORD undef %10, %1.sub2, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORD undef %10, %1.sub2, 0, 0, 0, implicit %exec, implicit %flat_scr
     %14 = COPY %1.sub1
     %16 = COPY killed %1.sub0
     undef %15.sub0 = COPY killed %16
     %15.sub1 = COPY killed %14
-    FLAT_STORE_DWORDX2 undef %11, killed %15, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORDX2 undef %11, killed %15, 0, 0, 0, implicit %exec, implicit %flat_scr
     S_ENDPGM
 
 ...

llvm/test/CodeGen/AMDGPU/sdwa-scalar-ops.mir

@@ -214,26 +214,26 @@ body:             |
     %15 = S_ADDC_U32 %7.sub1, %0.sub1, implicit-def dead %scc, implicit %scc
     %16 = REG_SEQUENCE %14, 1, %15, 2
     %18 = COPY %16
-    %17 = FLAT_LOAD_DWORD %18, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.uglygep45)
+    %17 = FLAT_LOAD_DWORD %18, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.uglygep45)
     %60 = V_BFE_U32 %17, 8, 8, implicit %exec
     %61 = V_LSHLREV_B32_e32 2, killed %60, implicit %exec
     %70 = V_ADD_I32_e32 %7.sub0, %61, implicit-def %vcc, implicit %exec
     %66 = COPY %13
     %65 = V_ADDC_U32_e32 0, %66, implicit-def %vcc, implicit %vcc, implicit %exec
     %67 = REG_SEQUENCE %70, 1, killed %65, 2
-    FLAT_STORE_DWORD %67, %30, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp9)
+    FLAT_STORE_DWORD %67, %30, 0, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp9)
     %37 = S_ADD_U32 %14, 4, implicit-def %scc
     %38 = S_ADDC_U32 %15, 0, implicit-def dead %scc, implicit %scc
     %71 = COPY killed %37
     %72 = COPY killed %38
     %41 = REG_SEQUENCE killed %71, 1, killed %72, 2
-    %40 = FLAT_LOAD_DWORD killed %41, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.scevgep)
+    %40 = FLAT_LOAD_DWORD killed %41, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.scevgep)
     %73 = V_BFE_U32 %40, 8, 8, implicit %exec
     %74 = V_LSHLREV_B32_e32 2, killed %73, implicit %exec
     %83 = V_ADD_I32_e32 %7.sub0, %74, implicit-def %vcc, implicit %exec
     %78 = V_ADDC_U32_e32 0, %66, implicit-def %vcc, implicit %vcc, implicit %exec
     %80 = REG_SEQUENCE %83, 1, killed %78, 2
-    FLAT_STORE_DWORD %80, %30, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp17)
+    FLAT_STORE_DWORD %80, %30, 0, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp17)
     %55 = S_ADD_U32 %0.sub0, 8, implicit-def %scc
     %56 = S_ADDC_U32 %0.sub1, 0, implicit-def dead %scc, implicit %scc
     %57 = REG_SEQUENCE %55, 1, killed %56, 2
@@ -377,26 +377,26 @@ body:             |
     %15 = S_ADDC_U32 %7.sub1, %0.sub1, implicit-def dead %scc, implicit %scc
     %16 = REG_SEQUENCE %14, 1, %15, 2
     %18 = COPY %16
-    %17 = FLAT_LOAD_DWORD %18, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.uglygep45)
+    %17 = FLAT_LOAD_DWORD %18, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.uglygep45)
     %60 = V_BFE_U32 %17, 8, 8, implicit %exec
     %61 = V_LSHLREV_B32_e32 %84, killed %60, implicit %exec
     %70 = V_ADD_I32_e32 %7.sub0, %61, implicit-def %vcc, implicit %exec
     %66 = COPY %13
     %65 = V_ADDC_U32_e32 0, %66, implicit-def %vcc, implicit %vcc, implicit %exec
     %67 = REG_SEQUENCE %70, 1, killed %65, 2
-    FLAT_STORE_DWORD %67, %30, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp9)
+    FLAT_STORE_DWORD %67, %30, 0, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp9)
     %37 = S_ADD_U32 %14, 4, implicit-def %scc
     %38 = S_ADDC_U32 %15, 0, implicit-def dead %scc, implicit %scc
     %71 = COPY killed %37
     %72 = COPY killed %38
     %41 = REG_SEQUENCE killed %71, 1, killed %72, 2
-    %40 = FLAT_LOAD_DWORD killed %41, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.scevgep)
+    %40 = FLAT_LOAD_DWORD killed %41, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.scevgep)
     %73 = V_BFE_U32 %40, 8, 8, implicit %exec
     %74 = V_LSHLREV_B32_e32 %84, killed %73, implicit %exec
     %83 = V_ADD_I32_e32 %7.sub0, %74, implicit-def %vcc, implicit %exec
     %78 = V_ADDC_U32_e32 0, %66, implicit-def %vcc, implicit %vcc, implicit %exec
     %80 = REG_SEQUENCE %83, 1, killed %78, 2
-    FLAT_STORE_DWORD %80, %30, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp17)
+    FLAT_STORE_DWORD %80, %30, 0, 0, 0, implicit %exec, implicit %flat_scr :: (store 4 into %ir.tmp17)
     %55 = S_ADD_U32 %0.sub0, 8, implicit-def %scc
     %56 = S_ADDC_U32 %0.sub1, 0, implicit-def dead %scc, implicit %scc
     %57 = REG_SEQUENCE %55, 1, killed %56, 2

llvm/test/CodeGen/AMDGPU/waitcnt.mir

@@ -51,21 +51,21 @@ name: flat_zero_waitcnt
 body: |
   bb.0:
     successors: %bb.1
-    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.global4)
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.global4)
-    %vgpr3_vgpr4_vgpr5_vgpr6 = FLAT_LOAD_DWORDX4 %vgpr7_vgpr8, 0, 0, implicit %exec, implicit %flat_scr :: (load 16 from %ir.global16)
+    %vgpr3_vgpr4_vgpr5_vgpr6 = FLAT_LOAD_DWORDX4 %vgpr7_vgpr8, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 16 from %ir.global16)
     %vgpr0 = V_MOV_B32_e32 %vgpr1, implicit %exec
     S_BRANCH %bb.1
 
   bb.1:
     successors: %bb.2
-    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, 0, implicit %exec, implicit %flat_scr
-    %vgpr3_vgpr4_vgpr5_vgpr6 = FLAT_LOAD_DWORDX4 %vgpr7_vgpr8, 0, 0, implicit %exec, implicit %flat_scr :: (load 16 from %ir.global16)
+    %vgpr3_vgpr4_vgpr5_vgpr6 = FLAT_LOAD_DWORDX4 %vgpr7_vgpr8, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 16 from %ir.global16)
     %vgpr0 = V_MOV_B32_e32 %vgpr1, implicit %exec
     S_BRANCH %bb.2
 
   bb.2:
-    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.flat4)
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4 from %ir.flat4)
-    %vgpr3_vgpr4_vgpr5_vgpr6 = FLAT_LOAD_DWORDX4 %vgpr7_vgpr8, 0, 0, implicit %exec, implicit %flat_scr :: (load 16 from %ir.flat16)
+    %vgpr3_vgpr4_vgpr5_vgpr6 = FLAT_LOAD_DWORDX4 %vgpr7_vgpr8, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 16 from %ir.flat16)
     %vgpr0 = V_MOV_B32_e32 %vgpr1, implicit %exec
     S_ENDPGM
 ...
@@ -86,11 +86,11 @@ name: single_fallthrough_successor_no_end_block_wait
 body: |
   bb.0:
     successors: %bb.1
-    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, 0, implicit %exec, implicit %flat_scr
 
   bb.1:
     %vgpr3_vgpr4 = V_LSHLREV_B64 4, %vgpr7_vgpr8, implicit %exec
-    FLAT_STORE_DWORD %vgpr3_vgpr4, %vgpr0, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORD %vgpr3_vgpr4, %vgpr0, 0, 0, 0, implicit %exec, implicit %flat_scr
     S_ENDPGM
 ...
 ---
@@ -114,15 +114,15 @@ name: single_branch_successor_not_next_block
 body: |
   bb.0:
     successors: %bb.2
-    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr1_vgpr2, 0, 0, 0, implicit %exec, implicit %flat_scr
    S_BRANCH %bb.2
 
   bb.1:
-    FLAT_STORE_DWORD %vgpr8_vgpr9, %vgpr10, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORD %vgpr8_vgpr9, %vgpr10, 0, 0, 0, implicit %exec, implicit %flat_scr
     S_ENDPGM
 
   bb.2:
      %vgpr3_vgpr4 = V_LSHLREV_B64 4, %vgpr7_vgpr8, implicit %exec
-    FLAT_STORE_DWORD %vgpr3_vgpr4, %vgpr0, 0, 0, implicit %exec, implicit %flat_scr
+    FLAT_STORE_DWORD %vgpr3_vgpr4, %vgpr0, 0, 0, 0, implicit %exec, implicit %flat_scr
     S_ENDPGM
 ...

llvm/test/MC/AMDGPU/flat-gfx9.s

@@ -0,0 +1,40 @@
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx900 -show-encoding %s | FileCheck -check-prefix=GFX9 -check-prefix=GCN %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=tonga -show-encoding %s | FileCheck -check-prefix=VI -check-prefix=GCN %s
+
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx900 -show-encoding 2>&1 %s | FileCheck -check-prefix=GFX9-ERR -check-prefix=GCNERR %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=tonga -show-encoding 2>&1 %s | FileCheck -check-prefix=VI-ERR -check-prefix=GCNERR %s
+
+
+flat_load_dword v1, v[3:4] offset:0
+// GCN: flat_load_dword v1, v[3:4]      ; encoding: [0x00,0x00,0x50,0xdc,0x03,0x00,0x00,0x01]
+
+flat_load_dword v1, v[3:4] offset:-1
+// GCN-ERR: :35: error: failed parsing operand.
+
+// FIXME: Error on VI in wrong column
+flat_load_dword v1, v[3:4] offset:4095
+// GFX9: flat_load_dword v1, v[3:4] offset:4095 ; encoding: [0xff,0x0f,0x50,0xdc,0x03,0x00,0x00,0x01]
+// VIERR: :1: error: invalid operand for instruction
+
+flat_load_dword v1, v[3:4] offset:4096
+// GCNERR: :28: error: invalid operand for instruction
+
+flat_load_dword v1, v[3:4] offset:4 glc
+// GFX9: flat_load_dword v1, v[3:4] offset:4 glc ; encoding: [0x04,0x00,0x51,0xdc,0x03,0x00,0x00,0x01]
+// VIERR: :1: error: invalid operand for instruction
+
+flat_load_dword v1, v[3:4] offset:4 glc slc
+// GFX9: flat_load_dword v1, v[3:4] offset:4 glc slc ; encoding: [0x04,0x00,0x53,0xdc,0x03,0x00,0x00,0x01]
+// VIERR: :1: error: invalid operand for instruction
+
+flat_atomic_add v[3:4], v5 offset:8 slc
+// GFX9: flat_atomic_add v[3:4], v5 offset:8 slc ; encoding: [0x08,0x00,0x0a,0xdd,0x03,0x05,0x00,0x00]
+// VIERR: :1: error: invalid operand for instruction
+
+flat_atomic_swap v[3:4], v5 offset:16
+// GFX9: flat_atomic_swap v[3:4], v5 offset:16 ; encoding: [0x10,0x00,0x00,0xdd,0x03,0x05,0x00,0x00]
+// VIERR: :1: error: invalid operand for instruction
+
+flat_store_dword v[3:4], v1 offset:16
+// GFX9: flat_store_dword v[3:4], v1 offset:16 ; encoding: [0x10,0x00,0x70,0xdc,0x03,0x01,0x00,0x00]
+// VIERR: :1: error: invalid operand for instruction

llvm/test/MC/AMDGPU/flat.s

@@ -49,9 +49,10 @@ flat_store_dword v[3:4], v1 slc
 
 // FIXME: For atomic instructions, glc must be placed immediately following
 // the data regiser.  These forms aren't currently supported:
+// FIXME: offset:0 required
 // flat_atomic_add v1, v[3:4], v5 slc glc
 
-flat_atomic_add v1 v[3:4], v5 glc slc
+flat_atomic_add v1, v[3:4], v5 offset:0 glc slc
 // NOSI: error:
 // CI: flat_atomic_add v1, v[3:4], v5 glc slc ; encoding: [0x00,0x00,0xcb,0xdc,0x03,0x05,0x00,0x01]
 // VI: flat_atomic_add v1, v[3:4], v5 glc slc ; encoding: [0x00,0x00,0x0b,0xdd,0x03,0x05,0x00,0x01]