llvm · jhuber6 · Feb 13, 2024 · Feb 9, 2024 · Artem-B · Feb 12, 2024
diff --git a/clang/docs/LanguageExtensions.rst b/clang/docs/LanguageExtensions.rst
@@ -2764,6 +2764,39 @@ Query for this feature with ``__has_builtin(__builtin_readcyclecounter)``. Note
 that even if present, its use may depend on run-time privilege or other OS
 controlled state.
 
+``__builtin_readsteadycounter``
+------------------------------
+
+``__builtin_readsteadycounter`` is used to access the fixed frequency counter
+register (or a similar steady-rate clock) on those targets that support it.
+The function is similar to ``__builtin_readcyclecounter`` above except that the
+frequency is fixed, making it suitable for measuring elapsed time.
+
+**Syntax**:
+
+.. code-block:: c++
+
+  __builtin_readsteadycounter()
+
+**Example of Use**:
+
+.. code-block:: c++
+
+  unsigned long long t0 = __builtin_readsteadycounter();
+  do_something();
+  unsigned long long t1 = __builtin_readsteadycounter();
+  unsigned long long secs_to_do_something = (t1 - t0) / tick_rate;
+
+**Description**:
+
+The ``__builtin_readsteadycounter()`` builtin returns the frequency counter value.
+When not supported by the target, the return value is always zero. This builtin
+takes no arguments and produces an unsigned long long result. The builtin does 
+not guarantee any particular frequency, only that it is stable. Knowledge of the 
+counter's true frequency will need to be provided by the user.
+
+Query for this feature with ``__has_builtin(__builtin_readsteadycounter)``.
+
 ``__builtin_dump_struct``
 -------------------------
 

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -117,6 +117,9 @@ C23 Feature Support
 Non-comprehensive list of changes in this release
 -------------------------------------------------
 
+- Added ``__builtin_readsteadycounter`` for reading fixed frequency hardware
+  counters.
+
 New Compiler Flags
 ------------------
 

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
@@ -1110,6 +1110,12 @@ def ReadCycleCounter : Builtin {
   let Prototype = "unsigned long long int()";
 }
 
+def ReadSteadyCounter : Builtin {
+  let Spellings = ["__builtin_readsteadycounter"];
+  let Attributes = [NoThrow];
+  let Prototype = "unsigned long long int()";
+}
+
 def Trap : Builtin {
   let Spellings = ["__builtin_trap"];
   let Attributes = [NoThrow, NoReturn];

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -3443,6 +3443,10 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
     Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter);
     return RValue::get(Builder.CreateCall(F));
   }
+  case Builtin::BI__builtin_readsteadycounter: {
+    Function *F = CGM.getIntrinsic(Intrinsic::readsteadycounter);
+    return RValue::get(Builder.CreateCall(F));
+  }
   case Builtin::BI__builtin___clear_cache: {
     Value *Begin = EmitScalarExpr(E->getArg(0));
     Value *End = EmitScalarExpr(E->getArg(1));

diff --git a/clang/test/CodeGen/builtins.c b/clang/test/CodeGen/builtins.c
@@ -496,6 +496,12 @@ long long test_builtin_readcyclecounter(void) {
   return __builtin_readcyclecounter();
 }
 
+// CHECK-LABEL: define{{.*}} i64 @test_builtin_readsteadycounter
+long long test_builtin_readsteadycounter(void) {
+  // CHECK: call i64 @llvm.readsteadycounter()
+  return __builtin_readsteadycounter();
+}
+
 /// __builtin_launder should be a NOP in C since there are no vtables.
 // CHECK-LABEL: define{{.*}} void @test_builtin_launder
 void test_builtin_launder(int *p) {

diff --git a/llvm/include/llvm/CodeGen/ISDOpcodes.h b/llvm/include/llvm/CodeGen/ISDOpcodes.h
@@ -1179,6 +1179,12 @@ enum NodeType {
   /// counter-like register (or other high accuracy low latency clock source).
   READCYCLECOUNTER,
 
+  /// READSTEADYCOUNTER - This corresponds to the readfixedcounter intrinsic.
+  /// It has the same semantics as the READCYCLECOUNTER implementation except
+  /// that the result is the content of the architecture-specific fixed
+  /// frequency counter suitable for measuring elapsed time.
+  READSTEADYCOUNTER,
+
   /// HANDLENODE node - Used as a handle for various purposes.
   HANDLENODE,
 

diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
@@ -870,6 +870,8 @@ def int_pcmarker      : DefaultAttrsIntrinsic<[], [llvm_i32_ty]>;
 
 def int_readcyclecounter : DefaultAttrsIntrinsic<[llvm_i64_ty]>;
 
+def int_readsteadycounter : DefaultAttrsIntrinsic<[llvm_i64_ty]>;
+
 // The assume intrinsic is marked InaccessibleMemOnly so that proper control
 // dependencies will be maintained.
 def int_assume : DefaultAttrsIntrinsic<

diff --git a/llvm/include/llvm/Support/TargetOpcodes.def b/llvm/include/llvm/Support/TargetOpcodes.def
@@ -352,6 +352,9 @@ HANDLE_TARGET_OPCODE(G_INTRINSIC_ROUNDEVEN)
 /// INTRINSIC readcyclecounter
 HANDLE_TARGET_OPCODE(G_READCYCLECOUNTER)
 
+/// INTRINSIC readsteadycounter
+HANDLE_TARGET_OPCODE(G_READSTEADYCOUNTER)
+
 /// Generic load (including anyext load)
 HANDLE_TARGET_OPCODE(G_LOAD)
 

diff --git a/llvm/include/llvm/Target/GenericOpcodes.td b/llvm/include/llvm/Target/GenericOpcodes.td
@@ -1101,6 +1101,12 @@ def G_READCYCLECOUNTER : GenericInstruction {
   let hasSideEffects = true;
 }
 
+def G_READSTEADYCOUNTER : GenericInstruction {
+  let OutOperandList = (outs type0:$dst);
+  let InOperandList = (ins);
+  let hasSideEffects = true;
+}
+
 //------------------------------------------------------------------------------
 // Memory ops
 //------------------------------------------------------------------------------

diff --git a/llvm/include/llvm/Target/GlobalISel/SelectionDAGCompat.td b/llvm/include/llvm/Target/GlobalISel/SelectionDAGCompat.td
@@ -168,6 +168,7 @@ def : GINodeEquiv<G_FMAXNUM_IEEE, fmaxnum_ieee>;
 def : GINodeEquiv<G_FMAXIMUM, fmaximum>;
 def : GINodeEquiv<G_FMINIMUM, fminimum>;
 def : GINodeEquiv<G_READCYCLECOUNTER, readcyclecounter>;
+def : GINodeEquiv<G_READSTEADYCOUNTER, readsteadycounter>;
 def : GINodeEquiv<G_ROTR, rotr>;
 def : GINodeEquiv<G_ROTL, rotl>;
 def : GINodeEquiv<G_LROUND, lround>;

diff --git a/llvm/include/llvm/Target/TargetSelectionDAG.td b/llvm/include/llvm/Target/TargetSelectionDAG.td
@@ -657,6 +657,9 @@ def prefetch   : SDNode<"ISD::PREFETCH"   , SDTPrefetch,
 def readcyclecounter : SDNode<"ISD::READCYCLECOUNTER", SDTIntLeaf,
                      [SDNPHasChain, SDNPSideEffect]>;
 
+def readsteadycounter : SDNode<"ISD::READSTEADYCOUNTER", SDTIntLeaf,
+                     [SDNPHasChain, SDNPSideEffect]>;
+
 def membarrier : SDNode<"ISD::MEMBARRIER", SDTNone,
                         [SDNPHasChain, SDNPSideEffect]>;
 

diff --git a/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp b/llvm/lib/CodeGen/GlobalISel/IRTranslator.cpp
@@ -1885,6 +1885,8 @@ unsigned IRTranslator::getSimpleIntrinsicOpcode(Intrinsic::ID ID) {
       return TargetOpcode::G_INTRINSIC_TRUNC;
     case Intrinsic::readcyclecounter:
       return TargetOpcode::G_READCYCLECOUNTER;
+    case Intrinsic::readsteadycounter:
+      return TargetOpcode::G_READSTEADYCOUNTER;
     case Intrinsic::ptrmask:
       return TargetOpcode::G_PTRMASK;
     case Intrinsic::lrint:

diff --git a/llvm/lib/CodeGen/IntrinsicLowering.cpp b/llvm/lib/CodeGen/IntrinsicLowering.cpp
@@ -312,6 +312,12 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
     CI->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context), 0));
     break;
   }
+  case Intrinsic::readsteadycounter: {
+    errs() << "WARNING: this target does not support the llvm.readsteadycounter"
+           << " intrinsic.  It is being lowered to a constant 0\n";
+    CI->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context), 0));
+    break;
+  }
 
   case Intrinsic::dbg_declare:
   case Intrinsic::dbg_label:

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -1127,8 +1127,9 @@ void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
       Action = TargetLowering::Custom;
     break;
   case ISD::READCYCLECOUNTER:
-    // READCYCLECOUNTER returns an i64, even if type legalization might have
-    // expanded that to several smaller types.
+  case ISD::READSTEADYCOUNTER:
+    // READCYCLECOUNTER and READSTEADYCOUNTER return a i64, even if type
+    // legalization might have expanded that to several smaller types.
     Action = TLI.getOperationAction(Node->getOpcode(), MVT::i64);
     break;
   case ISD::READ_REGISTER:
@@ -3080,6 +3081,7 @@ bool SelectionDAGLegalize::ExpandNode(SDNode *Node) {
     Results.push_back(Node->getOperand(0));
     break;
   case ISD::READCYCLECOUNTER:
+  case ISD::READSTEADYCOUNTER:
     // If the target didn't expand this, just return 'zero' and preserve the
     // chain.
     Results.append(Node->getNumValues() - 1,

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -2648,7 +2648,8 @@ void DAGTypeLegalizer::ExpandIntegerResult(SDNode *N, unsigned ResNo) {
   case ISD::LLRINT:      ExpandIntRes_XROUND_XRINT(N, Lo, Hi); break;
   case ISD::LOAD:        ExpandIntRes_LOAD(cast<LoadSDNode>(N), Lo, Hi); break;
   case ISD::MUL:         ExpandIntRes_MUL(N, Lo, Hi); break;
-  case ISD::READCYCLECOUNTER: ExpandIntRes_READCYCLECOUNTER(N, Lo, Hi); break;
+  case ISD::READCYCLECOUNTER:
+  case ISD::READSTEADYCOUNTER: ExpandIntRes_READCOUNTER(N, Lo, Hi); break;
   case ISD::SDIV:        ExpandIntRes_SDIV(N, Lo, Hi); break;
   case ISD::SIGN_EXTEND: ExpandIntRes_SIGN_EXTEND(N, Lo, Hi); break;
   case ISD::SIGN_EXTEND_INREG: ExpandIntRes_SIGN_EXTEND_INREG(N, Lo, Hi); break;
@@ -4026,8 +4027,8 @@ void DAGTypeLegalizer::ExpandIntRes_MUL(SDNode *N,
                Lo, Hi);
 }
 
-void DAGTypeLegalizer::ExpandIntRes_READCYCLECOUNTER(SDNode *N, SDValue &Lo,
-                                                     SDValue &Hi) {
+void DAGTypeLegalizer::ExpandIntRes_READCOUNTER(SDNode *N, SDValue &Lo,
+                                                SDValue &Hi) {
   SDLoc DL(N);
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDVTList VTs = DAG.getVTList(NVT, NVT, MVT::Other);

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -439,7 +439,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   void ExpandIntRes_CTPOP             (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_CTTZ              (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_LOAD          (LoadSDNode *N, SDValue &Lo, SDValue &Hi);
-  void ExpandIntRes_READCYCLECOUNTER  (SDNode *N, SDValue &Lo, SDValue &Hi);
+  void ExpandIntRes_READCOUNTER       (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_SIGN_EXTEND       (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_SIGN_EXTEND_INREG (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_TRUNCATE          (SDNode *N, SDValue &Lo, SDValue &Hi);

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -6781,6 +6781,14 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
     DAG.setRoot(Res.getValue(1));
     return;
   }
+  case Intrinsic::readsteadycounter: {
+    SDValue Op = getRoot();
+    Res = DAG.getNode(ISD::READSTEADYCOUNTER, sdl,
+                      DAG.getVTList(MVT::i64, MVT::Other), Op);
+    setValue(&I, Res);
+    DAG.setRoot(Res.getValue(1));
+    return;
+  }
   case Intrinsic::bitreverse:
     setValue(&I, DAG.getNode(ISD::BITREVERSE, sdl,
                              getValue(I.getArgOperand(0)).getValueType(),

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -104,6 +104,7 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
   case ISD::ATOMIC_STORE:               return "AtomicStore";
   case ISD::PCMARKER:                   return "PCMarker";
   case ISD::READCYCLECOUNTER:           return "ReadCycleCounter";
+  case ISD::READSTEADYCOUNTER:          return "ReadSteadyCounter";
   case ISD::SRCVALUE:                   return "SrcValue";
   case ISD::MDNODE_SDNODE:              return "MDNode";
   case ISD::EntryToken:                 return "EntryToken";

diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -964,6 +964,9 @@ void TargetLoweringBase::initActions() {
   // Most targets also ignore the @llvm.readcyclecounter intrinsic.
   setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, Expand);
 
+  // Most targets also ignore the @llvm.readsteadycounter intrinsic.
+  setOperationAction(ISD::READSTEADYCOUNTER, MVT::i64, Expand);
+
   // ConstantFP nodes default to expand.  Targets can either change this to
   // Legal, in which case all fp constants are legal, or use isFPImmLegal()
   // to optimize expansions for certain constants.

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -1957,6 +1957,8 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
   getActionDefinitionsBuilder(G_READCYCLECOUNTER)
     .legalFor({S64});
 
+  getActionDefinitionsBuilder(G_READSTEADYCOUNTER).legalFor({S64});
+
   getActionDefinitionsBuilder(G_FENCE)
     .alwaysLegal();
 

diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -4051,6 +4051,7 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
   case AMDGPU::G_CONSTANT:
   case AMDGPU::G_GLOBAL_VALUE:
   case AMDGPU::G_BLOCK_ADDR:
+  case AMDGPU::G_READSTEADYCOUNTER:
   case AMDGPU::G_READCYCLECOUNTER: {
     unsigned Size = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
     OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -468,6 +468,10 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
 
   // On SI this is s_memtime and s_memrealtime on VI.
   setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, Legal);
+
+  if (Subtarget->hasSMemRealTime() ||
+      Subtarget->getGeneration() >= AMDGPUSubtarget::GFX11)
+    setOperationAction(ISD::READSTEADYCOUNTER, MVT::i64, Legal);
   setOperationAction({ISD::TRAP, ISD::DEBUGTRAP}, MVT::Other, Custom);
 
   if (Subtarget->has16BitInsts()) {

diff --git a/llvm/lib/Target/AMDGPU/SMInstructions.td b/llvm/lib/Target/AMDGPU/SMInstructions.td
@@ -1068,6 +1068,20 @@ def : GCNPat <
 }
 } // let OtherPredicates = [HasShaderCyclesRegister]
 
+let OtherPredicates = [HasSMemRealTime] in {
+def : GCNPat <
+  (i64 (readsteadycounter)),
+  (S_MEMREALTIME)
+>;
+} // let OtherPredicates = [HasSMemRealTime]
+
+let SubtargetPredicate = isGFX11Plus in {
+def : GCNPat <
+  (i64 (readsteadycounter)),
+  (S_SENDMSG_RTN_B64 (i32 /*MSG_RTN_GET_REALTIME=*/0x83))
+>;
+} // let SubtargetPredicate = [isGFX11Plus]
+
 def i32imm_zero : TImmLeaf <i32, [{
   return Imm == 0;
 }]>;

diff --git a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -490,6 +490,8 @@ NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM,
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v2f16, Expand);
 
   setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, Legal);
+  if (STI.getSmVersion() >= 30 && STI.getPTXVersion() > 31)
+    setOperationAction(ISD::READSTEADYCOUNTER, MVT::i64, Legal);
 
   setFP16OperationAction(ISD::SETCC, MVT::f16, Legal, Promote);
   setFP16OperationAction(ISD::SETCC, MVT::v2f16, Legal, Expand);

diff --git a/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td b/llvm/lib/Target/NVPTX/NVPTXInstrInfo.td
@@ -3805,7 +3805,6 @@ def CALL_PROTOTYPE :
 
 include "NVPTXIntrinsics.td"
 
-
 //-----------------------------------
 // Notes
 //-----------------------------------

diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -6382,6 +6382,7 @@ def INT_PTX_SREG_GLOBALTIMER :
 }
 
 def: Pat <(i64 (readcyclecounter)), (INT_PTX_SREG_CLOCK64)>;
+def: Pat <(i64 (readsteadycounter)), (INT_PTX_SREG_GLOBALTIMER)>;
 
 def INT_PTX_SREG_PM0 : PTX_READ_SREG_R32<"pm0", int_nvvm_read_ptx_sreg_pm0>;
 def INT_PTX_SREG_PM1 : PTX_READ_SREG_R32<"pm1", int_nvvm_read_ptx_sreg_pm1>;

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
@@ -163,6 +163,9 @@
 # DEBUG-NEXT: G_READCYCLECOUNTER (opcode {{[0-9]+}}): 1 type index, 0 imm indices
 # DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: G_READSTEADYCOUNTER (opcode {{[0-9]+}}): 1 type index, 0 imm indices
+# DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
 
 # DEBUG-NEXT: G_LOAD (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
 # DEBUG-NEXT: .. type index coverage check SKIPPED: user-defined predicate detected

diff --git a/llvm/test/CodeGen/AMDGPU/readsteadycounter.ll b/llvm/test/CodeGen/AMDGPU/readsteadycounter.ll
@@ -0,0 +1,24 @@
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,GFX700
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,GFX900
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,GFX900
+; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,GFX1100
+; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,GFX1100
+
+declare i64 @llvm.readsteadycounter() #0
+
+; GCN-LABEL: {{^}}test_readsteadycounter:
+; GFX700: s_mov_b32 s[[REG:[0-9]+]], 0
+; GFX900: s_memrealtime s[[[LO:[0-9]+]]:[[HI:[0-9]+]]]
+; GFX900: s_memrealtime s[[[LO:[0-9]+]]:[[HI:[0-9]+]]]
+; GFX1100: s_sendmsg_rtn_b64 s[[[LO:[0-9]+]]:[[HI:[0-9]+]]], sendmsg(MSG_RTN_GET_REALTIME)
+; GFX1100: s_sendmsg_rtn_b64 s[[[LO:[0-9]+]]:[[HI:[0-9]+]]], sendmsg(MSG_RTN_GET_REALTIME)
+define amdgpu_kernel void @test_readsteadycounter(ptr addrspace(1) %out) #0 {
+  %cycle0 = call i64 @llvm.readsteadycounter()
+  store volatile i64 %cycle0, ptr addrspace(1) %out
+
+  %cycle1 = call i64 @llvm.readsteadycounter()
+  store volatile i64 %cycle1, ptr addrspace(1) %out
+  ret void
+}
+
+attributes #0 = { nounwind }
diff --git a/llvm/test/CodeGen/NVPTX/intrinsics.ll b/llvm/test/CodeGen/NVPTX/intrinsics.ll
@@ -162,6 +162,17 @@ define i64 @test_cyclecounter() {
   ret i64 %ret
 }
 
+; CHECK-LABEL: test_steadycounter
+define i64 @test_steadycounter() {
+; CHECK: mov.u64         %r{{.*}}, %globaltimer;
+  %a = tail call i64 @llvm.readsteadycounter()
+; CHECK: mov.u64         %r{{.*}}, %globaltimer;
+  %b = tail call i64 @llvm.readsteadycounter()
+  %ret = add i64 %a, %b
+; CHECK: ret
+  ret i64 %ret
+}
+
 declare float @llvm.fabs.f32(float)
 declare double @llvm.fabs.f64(double)
 declare float @llvm.nvvm.sqrt.f(float)
@@ -178,3 +189,4 @@ declare i64 @llvm.nvvm.read.ptx.sreg.clock64()
 declare void @llvm.nvvm.exit()
 declare i64 @llvm.nvvm.read.ptx.sreg.globaltimer()
 declare i64 @llvm.readcyclecounter()
+declare i64 @llvm.readsteadycounter()