[Clang - ARM/AArch64] Add ACLE special register intrinsics (10.1)

[LukeCheeseman]

Summary

This patch implements clang support for the ACLE special register intrinsics in section 10.1, __arm_{w,r}sr{,p,64}.

This includes arm_acle.h definitions with builtins and codegen to support these, the intrinsics are implemented by generating read/write_register calls which get appropriately lowered in the backend based on the register string provided. SemaChecking is also implemented to fault invalid parameters, however SemaChecking currently only supports checking in when the register string is in an encoded form, e.g. "cp1:2:c3:c4:5", that the encoding is valid. Otherwise, checking just assumes that the string is a special register value, e.g. "CPSR"; this means that if the register value passed is not a valid special register value, it passes checking and reports an error in the backend when attempting to lower the read/write_register intrinsic. I would like to have more checking on these values in the frontend to provide more useful diagnostics but couldn't think of a clean and nice way of doing it.

[rengolin]

lib/CodeGen/CGBuiltin.cpp

4285

This could be factored out into a static function together with the ARM counterpart above.

[LukeCheeseman]

Re-factored the code for the AArch64 and ARM special registers builtins to a common helper function.

[LukeCheeseman]

Updating patch for latests revision of sources.

[rengolin]

lib/CodeGen/CGBuiltin.cpp

3271

IsAArch64 in a generic function doesn't make sense. Is64Bit may make some sense, but I'd prefer if you pass what you really need.

In this case, I believe it's more about the types. So just receive what (input/output) types you need, and let the caller decide which types to pass.

4299

For example, you could use this boolean to create a type:

llvm::Type *Ty = (Is64Bit ? Int64Ty : Int32Ty);

and pass Ty, instead of the boolean.

lib/Sema/SemaChecking.cpp

2613

same argument here.

[LukeCheeseman]

Removing IsAArch64 (and Is64Bit) from helper functions to make them more generic. Replacing these parameters with the types for the operation when generating IR. Passing through the BuiltinID for the SemaChecking.

[rengolin]

Hi Luke,

Thanks for the change, looks good to me, now. Feel free to commit with the nitpick changes.

cheers,
--renato

lib/CodeGen/CGBuiltin.cpp

3297

nitpick: this could be commoned up with the write version below:

bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32);

Maybe we should assert that the other way around doesn't happen:

assert(!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && "Can't fit 64-bit value in 32-bit register");

[LukeCheeseman]

Hi Renato,

Thanks for the suggestions, I will apply them right away. I have one concern with respect to the semantic checking however, currently if an invalid register string is passed to one of the intrinsics the semantic checking won't detect this and so will not provide the user with a diagnostic. Instead, the error will occur in the backend when the write_register or read_register is to be lowered but cannot be done so successfully and a fatal error will occur (through report_fatal_error). For example:

$ cat tmp.c
#include <arm_acle.h>

unsigned read_notareg() {
  return __arm_rsr("notareg");
}

$ clang tmp.c -target arm-none-none-eabi -S
fatal error: error in backend: Invalid register name "notareg".
(error and bug report)

Is this acceptable? I believe those kinds of errors are for invalid user input. I couldn't think of way to check the possible strings in the semantic checking that would avoid bringing most of the logic into the frontend and so was wondering if you had any suggestions on the matter.

Thanks,
Luke

[LukeCheeseman]

ping

[rengolin]

In D9697#184235, @LukeCheeseman wrote:

Is this acceptable? I believe those kinds of errors are for invalid user input. I couldn't think of way to check the possible strings in the semantic checking that would avoid bringing most of the logic into the frontend and so was wondering if you had any suggestions on the matter.

Unfortunately, yes. There is currently no way of knowing what register names are available in Clang, which would need a Target Description that doesn't need all back-ends to be built (work in progress). For now, having that error in the back-end is the only way we can do, so it's acceptable.

cheers,
--renato

include/clang/Basic/BuiltinsAArch64.def

 // Memory barrier
 BUILTIN(__builtin_arm_dmb, "vUi", "nc")
 BUILTIN(__builtin_arm_dsb, "vUi", "nc")
 BUILTIN(__builtin_arm_isb, "vUi", "nc")
 
 // Prefetch
 BUILTIN(__builtin_arm_prefetch, "vvC*UiUiUiUi", "nc")
 
+// System Registers
+BUILTIN(__builtin_arm_rsr, "UicC*", "nc")
+BUILTIN(__builtin_arm_rsr64, "LUicC*", "nc")
+BUILTIN(__builtin_arm_rsrp, "v*cC*", "nc")
+BUILTIN(__builtin_arm_wsr, "vcC*Ui", "nc")
+BUILTIN(__builtin_arm_wsr64, "vcC*LUi", "nc")
+BUILTIN(__builtin_arm_wsrp, "vcC*vC*", "nc")
+
 #undef BUILTIN

include/clang/Basic/BuiltinsARM.def

 // Data barrier
 BUILTIN(__builtin_arm_dmb, "vUi", "nc")
 BUILTIN(__builtin_arm_dsb, "vUi", "nc")
 BUILTIN(__builtin_arm_isb, "vUi", "nc")
 
 // Prefetch
 BUILTIN(__builtin_arm_prefetch, "vvC*UiUi", "nc")
 
+// System registers (ACLE)
+BUILTIN(__builtin_arm_rsr, "UicC*", "nc")
+BUILTIN(__builtin_arm_rsr64, "LLUicC*", "nc")
+BUILTIN(__builtin_arm_rsrp, "v*cC*", "nc")
+BUILTIN(__builtin_arm_wsr, "vcC*Ui", "nc")
+BUILTIN(__builtin_arm_wsr64, "vcC*LLUi", "nc")
+BUILTIN(__builtin_arm_wsrp, "vcC*vC*", "nc")
+
 // MSVC
 LANGBUILTIN(__emit, "vIUiC", "", ALL_MS_LANGUAGES)
 
 LANGBUILTIN(__yield, "v", "", ALL_MS_LANGUAGES)
 LANGBUILTIN(__wfe, "v", "", ALL_MS_LANGUAGES)
 LANGBUILTIN(__wfi, "v", "", ALL_MS_LANGUAGES)
 LANGBUILTIN(__sev, "v", "", ALL_MS_LANGUAGES)
 LANGBUILTIN(__sevl, "v", "", ALL_MS_LANGUAGES)

include/clang/Basic/DiagnosticSemaKinds.td

 def note_ice_conversion_here : Note<
   "conversion to %select{integral|enumeration}0 type %1 declared here">;
 def err_ice_ambiguous_conversion : Error<
   "ambiguous conversion from type %0 to an integral or unscoped "
   "enumeration type">;
 def err_ice_too_large : Error<
   "integer constant expression evaluates to value %0 that cannot be "
   "represented in a %1-bit %select{signed|unsigned}2 integer type">;
+def err_expr_not_string_literal : Error<"expression is not a string literal">;
 
 // Semantic analysis of constant literals.
 def ext_predef_outside_function : Warning<
   "predefined identifier is only valid inside function">,
   InGroup<DiagGroup<"predefined-identifier-outside-function">>;
 def warn_float_overflow : Warning<
   "magnitude of floating-point constant too large for type %0; maximum is %1">,
    InGroup<LiteralRange>;

 def note_previous_builtin_declaration : Note<"%0 is a builtin with type %1">;
 def warn_implicit_decl_requires_sysheader : Warning<
   "declaration of built-in function '%1' requires inclusion of the header <%0>">,
   InGroup<BuiltinRequiresHeader>;
 def warn_redecl_library_builtin : Warning<
   "incompatible redeclaration of library function %0">,
   InGroup<DiagGroup<"incompatible-library-redeclaration">>;
 def err_builtin_definition : Error<"definition of builtin function %0">;
+def err_arm_invalid_specialreg : Error<"invalid special register for builtin">;
 def warn_builtin_unknown : Warning<"use of unknown builtin %0">,
   InGroup<ImplicitFunctionDeclare>, DefaultError;
 def warn_dyn_class_memaccess : Warning<
   "%select{destination for|source of|first operand of|second operand of}0 this "
   "%1 call is a pointer to %select{|class containing a }2dynamic class %3; "
   "vtable pointer will be %select{overwritten|copied|moved|compared}4">,
   InGroup<DiagGroup<"dynamic-class-memaccess">>;
 def note_bad_memaccess_silence : Note<

include/clang/Sema/Sema.h

private:
   bool SemaBuiltinSetjmp(CallExpr *TheCall);
   ExprResult SemaBuiltinAtomicOverloaded(ExprResult TheCallResult);
   ExprResult SemaAtomicOpsOverloaded(ExprResult TheCallResult,
                                      AtomicExpr::AtomicOp Op);
   bool SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
                               llvm::APSInt &Result);
   bool SemaBuiltinConstantArgRange(CallExpr *TheCall, int ArgNum,
                                    int Low, int High);
+  bool SemaBuiltinARMSpecialReg(CallExpr *TheCall, int ArgNum,
+                                unsigned ExpectedFieldNum, bool AllowName,
+                                bool isAArch64);
 
 public:
   enum FormatStringType {
     FST_Scanf,
     FST_Printf,
     FST_NSString,
     FST_Strftime,
     FST_Strfmon,

lib/CodeGen/CGBuiltin.cpp

Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) {
   case ARM::BI__sevl:
     return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint),
                               llvm::ConstantInt::get(Int32Ty, 5));
   }
 }
 
 Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
                                            const CallExpr *E) {
   if (auto Hint = GetValueForARMHint(BuiltinID))
     return Hint;
 
   if (BuiltinID == ARM::BI__emit) {
     bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb;
     llvm::FunctionType *FTy =
         llvm::FunctionType::get(VoidTy, /*Variadic=*/false);
 
     APSInt Value;
llvm::InlineAsm *Emit =
                                  /*SideEffects=*/true);
 
     return Builder.CreateCall(Emit);
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_dbg) {
     Value *Option = EmitScalarExpr(E->getArg(0));
     return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option);
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_prefetch) {
     Value *Address = EmitScalarExpr(E->getArg(0));
     Value *RW      = EmitScalarExpr(E->getArg(1));
     Value *IsData  = EmitScalarExpr(E->getArg(2));
 
     // Locality is not supported on ARM target
     Value *Locality = llvm::ConstantInt::get(Int32Ty, 3);
if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
     } else {
       Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty);
 
       Function *F = CGM.getIntrinsic(CRCIntrinsicID);
       return Builder.CreateCall2(F, Arg0, Arg1);
     }
   }
 
+  if (BuiltinID == ARM::BI__builtin_arm_rsr ||
+      BuiltinID == ARM::BI__builtin_arm_rsr64 ||
+      BuiltinID == ARM::BI__builtin_arm_rsrp ||
+      BuiltinID == ARM::BI__builtin_arm_wsr ||
+      BuiltinID == ARM::BI__builtin_arm_wsr64 ||
+      BuiltinID == ARM::BI__builtin_arm_wsrp) {
+
+    const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts();
+    StringRef SysReg = cast<StringLiteral>(SysRegStrExpr)->getString();
+    bool IsRead = BuiltinID == ARM::BI__builtin_arm_rsr ||
+                  BuiltinID == ARM::BI__builtin_arm_rsr64 ||
+                  BuiltinID == ARM::BI__builtin_arm_rsrp;
+    bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 ||
+                   BuiltinID == ARM::BI__builtin_arm_wsr64;
+
+    llvm::Metadata *Ops[] = {llvm::MDString::get(CGM.getLLVMContext(), SysReg)};
+    llvm::MDNode *RegName = llvm::MDNode::get(CGM.getLLVMContext(), Ops);
+    llvm::Value *Metadata =
+                      llvm::MetadataAsValue::get(CGM.getLLVMContext(), RegName);
+
+    llvm::Type *Types[] = { Is64Bit ? Int64Ty : Int32Ty };
+
+    if (IsRead) {
+      llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types);
+      llvm::Value *Call = Builder.CreateCall(F, Metadata);
+
+      if (BuiltinID == ARM::BI__builtin_arm_rsrp)
+        // Have i32 result (Call) but want to return a VoidPtrTy (i8*).
+        return Builder.CreateIntToPtr(Call, VoidPtrTy);
+
+      return Call;
+    }
+
+    llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
+
+    llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1));
+    if (BuiltinID == ARM::BI__builtin_arm_wsrp)
+      // Have ptr argument but want to pass i32 value to write.
+      ArgValue = Builder.CreatePtrToInt(ArgValue, Int32Ty);
+
+    return Builder.CreateCall2(F, Metadata, ArgValue);
+  }
+
   SmallVector<Value*, 4> Ops;
   llvm::Value *Align = nullptr;
   for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) {
     if (i == 0) {
       switch (BuiltinID) {
       case NEON::BI__builtin_neon_vld1_v:
       case NEON::BI__builtin_neon_vld1q_v:
       case NEON::BI__builtin_neon_vld1q_lane_v:
if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
     Function *F = CGM.getIntrinsic(CRCIntrinsicID);
 
     llvm::Type *DataTy = F->getFunctionType()->getParamType(1);
     Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy);
 
     return Builder.CreateCall2(F, Arg0, Arg1);
   }
 
+  if (BuiltinID == AArch64::BI__builtin_arm_rsr ||
+      BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
+      BuiltinID == AArch64::BI__builtin_arm_rsrp ||
+      BuiltinID == AArch64::BI__builtin_arm_wsr ||
+      BuiltinID == AArch64::BI__builtin_arm_wsr64 ||
+      BuiltinID == AArch64::BI__builtin_arm_wsrp) {
+    const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts();
+    StringRef SysReg = cast<StringLiteral>(SysRegStrExpr)->getString();
+
+    bool IsRead = BuiltinID == AArch64::BI__builtin_arm_rsr ||
+                  BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
+                  BuiltinID == AArch64::BI__builtin_arm_rsrp;
+
+    llvm::Metadata *Ops[] = {llvm::MDString::get(CGM.getLLVMContext(), SysReg)};
+    llvm::MDNode *RegName = llvm::MDNode::get(CGM.getLLVMContext(), Ops);
+    llvm::Value* Metadata =
+                      llvm::MetadataAsValue::get(CGM.getLLVMContext(), RegName);
+    llvm::Type *Types[] = { Int64Ty };
+
+    if (IsRead) {
+      llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types);
+      llvm::Value *Call = Builder.CreateCall(F, Metadata);
+
+      if (BuiltinID == AArch64::BI__builtin_arm_rsr)
+        // Read into 64 bit register via read register
+        // and then truncate result to 32 bit.
+        Call = Builder.CreateTrunc(Call, Int32Ty);
+
+      if (BuiltinID == AArch64::BI__builtin_arm_rsrp)
+        // Have i64 result (Call) but want to return a VoidPtrTy (i8*).
+        return Builder.CreateIntToPtr(Call, VoidPtrTy);
+
+      return Call;
+    }
+
+    llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
+    llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1));
+
+    if (BuiltinID == AArch64::BI__builtin_arm_wsr)
+      // Extend 32 bit write value to 64 bit to pass to write.
+      ArgValue = Builder.CreateZExt(ArgValue, Int64Ty);
+
+    if (BuiltinID == AArch64::BI__builtin_arm_wsrp)
+      // Have ptr argument but want to pass i64 value to write.
+      ArgValue = Builder.CreatePtrToInt(ArgValue, Int64Ty);
+
+    return Builder.CreateCall2(F, Metadata, ArgValue);
+  }
+
   llvm::SmallVector<Value*, 4> Ops;
   for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++)
     Ops.push_back(EmitScalarExpr(E->getArg(i)));
 
   auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap);
   const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap(
       SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted);
 

lib/Headers/arm_acle.h

 }
 
 static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32cd(uint32_t a, uint64_t b) {
   return __builtin_arm_crc32cd(a, b);
 }
 #endif
 
+/* 10.1 Special register intrinsics */
+#define __arm_rsr(sysreg) __builtin_arm_rsr(sysreg)
+#define __arm_rsr64(sysreg) __builtin_arm_rsr64(sysreg)
+#define __arm_rsrp(sysreg) __builtin_arm_rsrp(sysreg)
+#define __arm_wsr(sysreg, v) __builtin_arm_wsr(sysreg, v)
+#define __arm_wsr64(sysreg, v) __builtin_arm_wsr64(sysreg, v)
+#define __arm_wsrp(sysreg, v) __builtin_arm_wsrp(sysreg, v)
+
 #if defined(__cplusplus)
 }
 #endif
 
 #endif /* __ARM_ACLE_H */

lib/Sema/SemaChecking.cpp

if (BuiltinID == ARM::BI__builtin_arm_ldrex ||
     return CheckARMBuiltinExclusiveCall(BuiltinID, TheCall, 64);
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_prefetch) {
     return SemaBuiltinConstantArgRange(TheCall, 1, 0, 1) ||
       SemaBuiltinConstantArgRange(TheCall, 2, 0, 1);
   }
 
+  if (BuiltinID == ARM::BI__builtin_arm_rsr64 ||
+      BuiltinID == ARM::BI__builtin_arm_wsr64)
+    return SemaBuiltinARMSpecialReg(TheCall, 0, 3, false, false);
+
+  if (BuiltinID == ARM::BI__builtin_arm_rsr ||
+      BuiltinID == ARM::BI__builtin_arm_rsrp ||
+      BuiltinID == ARM::BI__builtin_arm_wsr ||
+      BuiltinID == ARM::BI__builtin_arm_wsrp)
+    return SemaBuiltinARMSpecialReg(TheCall, 0, 5, true, false);
+
   if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall))
     return true;
 
   // For intrinsics which take an immediate value as part of the instruction,
   // range check them here.
   unsigned i = 0, l = 0, u = 0;
   switch (BuiltinID) {
   default: return false;
bool Sema::CheckAArch64BuiltinFunctionCall(unsigned BuiltinID,
 
   if (BuiltinID == AArch64::BI__builtin_arm_prefetch) {
     return SemaBuiltinConstantArgRange(TheCall, 1, 0, 1) ||
       SemaBuiltinConstantArgRange(TheCall, 2, 0, 2) ||
       SemaBuiltinConstantArgRange(TheCall, 3, 0, 1) ||
       SemaBuiltinConstantArgRange(TheCall, 4, 0, 1);
   }
 
+  if (BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
+      BuiltinID == AArch64::BI__builtin_arm_wsr64)
+    return SemaBuiltinARMSpecialReg(TheCall, 0, 5, false, true);
+
+  if (BuiltinID == AArch64::BI__builtin_arm_rsr ||
+      BuiltinID == AArch64::BI__builtin_arm_rsrp ||
+      BuiltinID == AArch64::BI__builtin_arm_wsr ||
+      BuiltinID == AArch64::BI__builtin_arm_wsrp)
+    return SemaBuiltinARMSpecialReg(TheCall, 0, 5, true, true);
+
   if (CheckNeonBuiltinFunctionCall(BuiltinID, TheCall))
     return true;
 
   // For intrinsics which take an immediate value as part of the instruction,
   // range check them here.
   unsigned i = 0, l = 0, u = 0;
   switch (BuiltinID) {
   default: return false;
bool Sema::SemaBuiltinConstantArgRange(CallExpr *TheCall, int ArgNum,
 
   if (Result.getSExtValue() < Low || Result.getSExtValue() > High)
     return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
       << Low << High << Arg->getSourceRange();
 
   return false;
 }
 
+/// SemaBuiltinARMSpecialReg - Handle a check if argument ArgNum of CallExpr
+/// TheCall is an ARM/AArch64 special register string literal.
+bool Sema::SemaBuiltinARMSpecialReg(CallExpr *TheCall, int ArgNum,
+                                    unsigned ExpectedFieldNum, bool AllowName,
+                                    bool IsAArch64) {
+  // We can't check the value of a dependent argument.
+  Expr *Arg = TheCall->getArg(ArgNum);
+  if (Arg->isTypeDependent() || Arg->isValueDependent())
+    return false;
+
+  // Check if the argument is a string literal.
+  if (!isa<StringLiteral>(Arg->IgnoreParenImpCasts()))
+    return Diag(TheCall->getLocStart(), diag::err_expr_not_string_literal)
+           << Arg->getSourceRange();
+
+  // Check the type of special register given.
+  StringRef Reg = cast<StringLiteral>(Arg->IgnoreParenImpCasts())->getString();
+  SmallVector<StringRef, 6> Fields;
+  Reg.split(Fields, ":");
+
+  if (Fields.size() != ExpectedFieldNum && !(AllowName && Fields.size() == 1))
+    return Diag(TheCall->getLocStart(), diag::err_arm_invalid_specialreg)
+           << Arg->getSourceRange();
+
+  // If the string is the name of a register then we cannot check that it is
+  // valid here but if the string is of one the forms described in ACLE then we
+  // can check that the supplied fields are integers and within the valid
+  // ranges.
+  if (Fields.size() > 1) {
+    bool FiveFields = Fields.size() == 5;
+
+    bool ValidString = true;
+    if (!IsAArch64) {
+      ValidString &= Fields[0].startswith_lower("cp") ||
+                     Fields[0].startswith_lower("p");
+      if (ValidString)
+        Fields[0] =
+          Fields[0].drop_front(Fields[0].startswith_lower("cp") ? 2 : 1);
+
+      ValidString &= Fields[2].startswith_lower("c");
+      if (ValidString)
+        Fields[2] = Fields[2].drop_front(1);
+
+      if (FiveFields) {
+        ValidString &= Fields[3].startswith_lower("c");
+        if (ValidString)
+          Fields[3] = Fields[3].drop_front(1);
+      }
+    }
+
+    SmallVector<int, 5> Ranges = {15, 7, 15};
+    if (FiveFields)
+      Ranges = {IsAArch64 ? 1 : 15, 7, 7, 15, 15};
+
+    for (unsigned i=0; i<Fields.size(); ++i) {
+      int IntField;
+      ValidString &= !Fields[i].getAsInteger(10, IntField);
+      ValidString &= (IntField >= 0 && IntField <= Ranges[i]);
+    }
+
+    if (!ValidString)
+      return Diag(TheCall->getLocStart(), diag::err_arm_invalid_specialreg)
+             << Arg->getSourceRange();
+  }
+
+  if (IsAArch64 && Fields.size() == 1) {
+    // If the register name is one of those that appear in the condition below
+    // and the special register builtin being used is one of the write builtins,
+    // then we require that the argument provided for writing to the register
+    // is an integer constant expression. This is because it will be lowered to
+    // an MSR (immediate) instruction, so we need to know the immediate at
+    // compile time.
+    if (TheCall->getNumArgs() != 2)
+      return false;
+
+    std::string RegLower = Reg.lower();
+    if (RegLower != "spsel" && RegLower != "daifset" && RegLower != "daifclr" &&
+        RegLower != "pan" && RegLower != "uao")
+      return false;
+
+    return SemaBuiltinConstantArgRange(TheCall, 1, 0, 15);
+  }
+
+  return false;
+}
+
 /// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
 /// This checks that the target supports __builtin_longjmp and
 /// that val is a constant 1.
 bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
   if (!Context.getTargetInfo().hasSjLjLowering())
     return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_unsupported)
              << SourceRange(TheCall->getLocStart(), TheCall->getLocEnd());
 

test/CodeGen/arm_acle.c

 
 // ARM-LABEL: test_crc32cd
 // AArch32: call i32 @llvm.arm.crc32cw
 // AArch32: call i32 @llvm.arm.crc32cw
 // AArch64: call i32 @llvm.aarch64.crc32cx
 uint32_t test_crc32cd(uint32_t a, uint64_t b) {
   return __crc32cd(a, b);
 }
+
+/* 10.1 Special register intrinsics */
+// ARM-LABEL: test_rsr
+// AArch64: call i64 @llvm.read_register.i64(metadata !3)
+// AArch32: call i32 @llvm.read_register.i32(metadata !3)
+uint32_t test_rsr() {
+#ifdef __ARM_32BIT_STATE
+  return __arm_rsr("cp1:2:c3:c4:5");
+#else
+  return __arm_rsr("1:2:3:4:5");
+#endif
+}
+
+// ARM-LABEL: test_rsr64
+// AArch64: call i64 @llvm.read_register.i64(metadata !3)
+// AArch32: call i64 @llvm.read_register.i64(metadata !4)
+uint64_t test_rsr64() {
+#ifdef __ARM_32BIT_STATE
+  return __arm_rsr64("cp1:2:c3");
+#else
+  return __arm_rsr64("1:2:3:4:5");
+#endif
+}
+
+// ARM-LABEL: test_rsrp
+// AArch64: call i64 @llvm.read_register.i64(metadata !4)
+// AArch32: call i32 @llvm.read_register.i32(metadata !5)
+void *test_rsrp() {
+  return __arm_rsrp("sysreg");
+}
+
+// ARM-LABEL: test_wsr
+// AArch64: call void @llvm.write_register.i64(metadata !3, i64 %0)
+// AArch32: call void @llvm.write_register.i32(metadata !3, i32 %v)
+void test_wsr(uint32_t v) {
+#ifdef __ARM_32BIT_STATE
+  __arm_wsr("cp1:2:c3:c4:5", v);
+#else
+  __arm_wsr("1:2:3:4:5", v);
+#endif
+}
+
+// ARM-LABEL: test_wsr64
+// AArch64: call void @llvm.write_register.i64(metadata !3, i64 %v)
+// AArch32: call void @llvm.write_register.i64(metadata !4, i64 %v)
+void test_wsr64(uint64_t v) {
+#ifdef __ARM_32BIT_STATE
+  __arm_wsr64("cp1:2:c3", v);
+#else
+  __arm_wsr64("1:2:3:4:5", v);
+#endif
+}
+
+// ARM-LABEL: test_wsrp
+// AArch64: call void @llvm.write_register.i64(metadata !4, i64 %0)
+// AArch32: call void @llvm.write_register.i32(metadata !5, i32 %0)
+void test_wsrp(void *v) {
+  __arm_wsrp("sysreg", v);
+}
+
+// AArch32: !3 = !{!"cp1:2:c3:c4:5"}
+// AArch32: !4 = !{!"cp1:2:c3"}
+// AArch32: !5 = !{!"sysreg"}
+
+// AArch64: !3 = !{!"1:2:3:4:5"}
+// AArch64: !4 = !{!"sysreg"}

test/CodeGen/builtins-arm.c

// CHECK: call {{.*}} @llvm.prefetch(i8* %{{.*}}, i32 0, i32 3, i32 1)
 
   __builtin_arm_prefetch(&i, 1, 1);
 // CHECK: call {{.*}} @llvm.prefetch(i8* %{{.*}}, i32 1, i32 3, i32 1)
 
 
   __builtin_arm_prefetch(&i, 1, 0);
 // CHECK: call {{.*}} @llvm.prefetch(i8* %{{.*}}, i32 1, i32 3, i32 0)
 }
+
+unsigned rsr() {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = {{.*}} call i32 @llvm.read_register.i32(metadata !7)
+  // CHECK-NEXT: ret i32 [[V0]]
+  return __builtin_arm_rsr("cp1:2:c3:c4:5");
+}
+
+unsigned long long rsr64() {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = {{.*}} call i64 @llvm.read_register.i64(metadata !8)
+  // CHECK-NEXT: ret i64 [[V0]]
+  return __builtin_arm_rsr64("cp1:2:c3");
+}
+
+void *rsrp() {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = {{.*}} call i32 @llvm.read_register.i32(metadata !9)
+  // CHECK-NEXT: [[V1:[%A-Za-z0-9.]+]] = inttoptr i32 [[V0]] to i8*
+  // CHECK-NEXT: ret i8* [[V1]]
+  return __builtin_arm_rsrp("sysreg");
+}
+
+void wsr(unsigned v) {
+  // CHECK: call void @llvm.write_register.i32(metadata !7, i32 %v)
+  __builtin_arm_wsr("cp1:2:c3:c4:5", v);
+}
+
+void wsr64(unsigned long long v) {
+  // CHECK: call void @llvm.write_register.i64(metadata !8, i64 %v)
+  __builtin_arm_wsr64("cp1:2:c3", v);
+}
+
+void wsrp(void *v) {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = ptrtoint i8* %v to i32
+  // CHECK-NEXT: call void @llvm.write_register.i32(metadata !9, i32 [[V0]])
+  __builtin_arm_wsrp("sysreg", v);
+}
+
+// CHECK: !7 = !{!"cp1:2:c3:c4:5"}
+// CHECK: !8 = !{!"cp1:2:c3"}
+// CHECK: !9 = !{!"sysreg"}

test/CodeGen/builtins-arm64.c

 // CHECK: call {{.*}} @llvm.prefetch(i8* null, i32 0, i32 0, i32 1)
 
   __builtin_arm_prefetch(0, 0, 0, 1, 1); // pldl1strm
 // CHECK: call {{.*}} @llvm.prefetch(i8* null, i32 0, i32 0, i32 1)
 
   __builtin_arm_prefetch(0, 0, 0, 0, 0); // plil1keep
 // CHECK: call {{.*}} @llvm.prefetch(i8* null, i32 0, i32 3, i32 0)
 }
+
+unsigned rsr() {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = {{.*}} call i64 @llvm.read_register.i64(metadata !3)
+  // CHECK-NEXT: trunc i64 [[V0]] to i32
+  return __builtin_arm_rsr("1:2:3:4:5");
+}
+
+unsigned long rsr64() {
+  // CHECK: call i64 @llvm.read_register.i64(metadata !3)
+  return __builtin_arm_rsr64("1:2:3:4:5");
+}
+
+void *rsrp() {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = {{.*}} call i64 @llvm.read_register.i64(metadata !3)
+  // CHECK-NEXT: inttoptr i64 [[V0]] to i8*
+  return __builtin_arm_rsrp("1:2:3:4:5");
+}
+
+void wsr(unsigned v) {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = zext i32 %v to i64
+  // CHECK-NEXT: call void @llvm.write_register.i64(metadata !3, i64 [[V0]])
+  __builtin_arm_wsr("1:2:3:4:5", v);
+}
+
+void wsr64(unsigned long v) {
+  // CHECK: call void @llvm.write_register.i64(metadata !3, i64 %v)
+  __builtin_arm_wsr64("1:2:3:4:5", v);
+}
+
+void wsrp(void *v) {
+  // CHECK: [[V0:[%A-Za-z0-9.]+]] = ptrtoint i8* %v to i64
+  // CHECK-NEXT: call void @llvm.write_register.i64(metadata !3, i64 [[V0]])
+  __builtin_arm_wsrp("1:2:3:4:5", v);
+}
+
+// CHECK: !3 = !{!"1:2:3:4:5"}

test/Sema/aarch64-special-register.c

+// RUN: %clang_cc1 -ffreestanding -fsyntax-only -verify -triple aarch64 %s
+
+void string_literal(unsigned v) {
+  __builtin_arm_wsr(0, v); // expected-error {{expression is not a string literal}}
+}
+
+void wsr_1(unsigned v) {
+  __builtin_arm_wsr("sysreg", v);
+}
+
+void wsrp_1(void *v) {
+  __builtin_arm_wsrp("sysreg", v);
+}
+
+void wsr64_1(unsigned long v) {
+  __builtin_arm_wsr64("sysreg", v); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned rsr_1() {
+  return __builtin_arm_rsr("sysreg");
+}
+
+void *rsrp_1() {
+  return __builtin_arm_rsrp("sysreg");
+}
+
+unsigned long rsr64_1() {
+  return __builtin_arm_rsr64("sysreg"); //expected-error {{invalid special register for builtin}}
+}
+
+void wsr_2(unsigned v) {
+  __builtin_arm_wsr("0:1:2:3:4", v);
+}
+
+void wsrp_2(void *v) {
+  __builtin_arm_wsrp("0:1:2:3:4", v);
+}
+
+void wsr64_2(unsigned long v) {
+  __builtin_arm_wsr64("0:1:2:3:4", v);
+}
+
+unsigned rsr_2() {
+  return __builtin_arm_rsr("0:1:2:3:4");
+}
+
+void *rsrp_2() {
+  return __builtin_arm_rsrp("0:1:2:3:4");
+}
+
+unsigned long rsr64_2() {
+  return __builtin_arm_rsr64("0:1:2:3:4");
+}
+
+void wsr_3(unsigned v) {
+  __builtin_arm_wsr("0:1:2", v); //expected-error {{invalid special register for builtin}}
+}
+
+void wsrp_3(void *v) {
+  __builtin_arm_wsrp("0:1:2", v); //expected-error {{invalid special register for builtin}}
+}
+
+void wsr64_3(unsigned long v) {
+  __builtin_arm_wsr64("0:1:2", v); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned rsr_3() {
+  return __builtin_arm_rsr("0:1:2"); //expected-error {{invalid special register for builtin}}
+}
+
+void *rsrp_3() {
+  return __builtin_arm_rsrp("0:1:2"); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned long rsr64_3() {
+  return __builtin_arm_rsr64("0:1:2"); //expected-error {{invalid special register for builtin}}
+}

test/Sema/arm-special-register.c

+// RUN: %clang_cc1 -ffreestanding -fsyntax-only -verify -triple arm %s
+
+void string_literal(unsigned v) {
+  __builtin_arm_wsr(0, v); // expected-error {{expression is not a string literal}}
+}
+
+void wsr_1(unsigned v) {
+  __builtin_arm_wsr("sysreg", v);
+}
+
+void wsrp_1(void *v) {
+  __builtin_arm_wsrp("sysreg", v);
+}
+
+void wsr64_1(unsigned long v) {
+  __builtin_arm_wsr64("sysreg", v); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned rsr_1() {
+  return __builtin_arm_rsr("sysreg");
+}
+
+void *rsrp_1() {
+  return __builtin_arm_rsrp("sysreg");
+}
+
+unsigned long rsr64_1() {
+  return __builtin_arm_rsr64("sysreg"); //expected-error {{invalid special register for builtin}}
+}
+
+void wsr_2(unsigned v) {
+  __builtin_arm_wsr("cp0:1:c2:c3:4", v);
+}
+
+void wsrp_2(void *v) {
+  __builtin_arm_wsrp("cp0:1:c2:c3:4", v);
+}
+
+void wsr64_2(unsigned long v) {
+  __builtin_arm_wsr64("cp0:1:c2:c3:4", v); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned rsr_2() {
+  return __builtin_arm_rsr("cp0:1:c2:c3:4");
+}
+
+void *rsrp_2() {
+  return __builtin_arm_rsrp("cp0:1:c2:c3:4");
+}
+
+unsigned long rsr64_2() {
+  return __builtin_arm_rsr64("cp0:1:c2:c3:4"); //expected-error {{invalid special register for builtin}}
+}
+
+void wsr_3(unsigned v) {
+  __builtin_arm_wsr("cp0:1:c2", v); //expected-error {{invalid special register for builtin}}
+}
+
+void wsrp_3(void *v) {
+  __builtin_arm_wsrp("cp0:1:c2", v); //expected-error {{invalid special register for builtin}}
+}
+
+void wsr64_3(unsigned long v) {
+  __builtin_arm_wsr64("cp0:1:c2", v);
+}
+
+unsigned rsr_3() {
+  return __builtin_arm_rsr("cp0:1:c2"); //expected-error {{invalid special register for builtin}}
+}
+
+void *rsrp_3() {
+  return __builtin_arm_rsrp("cp0:1:c2"); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned long rsr64_3() {
+  return __builtin_arm_rsr64("cp0:1:c2");
+}
+
+unsigned rsr_4() {
+  return __builtin_arm_rsr("0:1:2:3:4"); //expected-error {{invalid special register for builtin}}
+}
+
+void *rsrp_4() {
+  return __builtin_arm_rsrp("0:1:2:3:4"); //expected-error {{invalid special register for builtin}}
+}
+
+unsigned long rsr64_4() {
+  return __builtin_arm_rsr64("0:1:2"); //expected-error {{invalid special register for builtin}}
+}