[MSVC, ARM64] Add __readx18 intrinsics

[steplong]

Summary

https://docs.microsoft.com/en-us/cpp/intrinsics/arm64-intrinsics?view=msvc-170

unsigned char __readx18byte(unsigned long)
unsigned short __readx18word(unsigned long)
unsigned long __readx18dword(unsigned long)
unsigned __int64 __readx18qword(unsigned long)

Given the lack of documentation of the intrinsics, we chose to align the offset with just
CharUnits::One() when calling IRBuilderBase::CreateAlignedLoad()

[steplong]

Like D126023, the generated assembly for MSVC is slightly different:

MSVC                                                                          LLVM
__readx18byte:
                          ldrb    w0, [x18, x0]                             ldrb    w0, [x18, w0, uxtw] 
__readx18word:
                          ldrh    w0, [x18, x0]                             ldrh    w0, [x18, w0, uxtw #1]
__readx18dword:
                          ldr     w0, [x18, x0]                             ldr     w0, [x18, w0, uxtw #2]
__readx18qword:
                          ldr     x0, [x18, x0]                              ldr     x0, [x18, w0, uxtw #3]

[steplong]

The assembly for LLVM is now:

__readx18byte:
       ldrb    w0, [x18, w0, uxtw]
__readx18word:
       ldrh    w0, [x18, w0, uxtw]
__readx18dword:
       ldr     w0, [x18, w0, uxtw]
__readx18qword:
       ldr     x0, [x18, w0, uxtw]

[steplong]

• Change addrspace(256) to the default addrspace 0

[steplong]

• Use CharUnits::One() instead of getTypeAlignInChars()

[efriedma]

LGTM

[steplong]

• Rebased patch on top of main

clang/include/clang/Basic/BuiltinsAArch64.def

 TARGET_HEADER_BUILTIN(_WriteStatusReg, "viLLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 TARGET_HEADER_BUILTIN(_AddressOfReturnAddress, "v*", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 
 TARGET_HEADER_BUILTIN(__mulh,  "SLLiSLLiSLLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 TARGET_HEADER_BUILTIN(__umulh, "ULLiULLiULLi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 
 TARGET_HEADER_BUILTIN(__break, "vi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
 
+TARGET_HEADER_BUILTIN(__readx18byte,  "UcULi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(__readx18word,  "UsULi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(__readx18dword, "ULiULi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(__readx18qword, "ULLiULi", "nh", "intrin.h", ALL_MS_LANGUAGES, "")
+
 #undef BUILTIN
 #undef LANGBUILTIN
 #undef TARGET_HEADER_BUILTIN

clang/lib/CodeGen/CGBuiltin.cpp

if (IsSigned) {
       MulResult = Builder.CreateNUWMul(LHS, RHS);
       HigherBits = Builder.CreateLShr(MulResult, 64);
     }
     HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned);
 
     return HigherBits;
   }
 
+  if (BuiltinID == AArch64::BI__readx18byte ||
+      BuiltinID == AArch64::BI__readx18word ||
+      BuiltinID == AArch64::BI__readx18dword ||
+      BuiltinID == AArch64::BI__readx18qword) {
+    llvm::Type *IntTy = ConvertType(E->getType());
+
+    // Read x18 as i8*
+    LLVMContext &Context = CGM.getLLVMContext();
+    llvm::Metadata *Ops[] = {llvm::MDString::get(Context, "x18")};
+    llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
+    llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
+    llvm::Function *F =
+        CGM.getIntrinsic(llvm::Intrinsic::read_register, {Int64Ty});
+    llvm::Value *X18 = Builder.CreateCall(F, Metadata);
+    X18 = Builder.CreateIntToPtr(X18, llvm::PointerType::get(Int8Ty, 0));
+
+    // Load x18 + offset
+    Value *Offset = Builder.CreateZExt(EmitScalarExpr(E->getArg(0)), Int64Ty);
+    Value *Ptr = Builder.CreateGEP(Int8Ty, X18, Offset);
+    Ptr = Builder.CreatePointerCast(Ptr, llvm::PointerType::get(IntTy, 0));
+    LoadInst *Load = Builder.CreateAlignedLoad(
+        IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
+    return Load;
+  }
+
   // Handle MSVC intrinsics before argument evaluation to prevent double
   // evaluation.
   if (Optional<MSVCIntrin> MsvcIntId = translateAarch64ToMsvcIntrin(BuiltinID))
     return EmitMSVCBuiltinExpr(*MsvcIntId, E);
 
   // Find out if any arguments are required to be integer constant
   // expressions.
   unsigned ICEArguments = 0;

clang/lib/Headers/intrin.h

 unsigned short __cdecl _byteswap_ushort(unsigned short val);
 unsigned long __cdecl _byteswap_ulong (unsigned long val);
 unsigned __int64 __cdecl _byteswap_uint64(unsigned __int64 val);
 
 __int64 __mulh(__int64 __a, __int64 __b);
 unsigned __int64 __umulh(unsigned __int64 __a, unsigned __int64 __b);
 
 void __break(int);
+
+unsigned char __readx18byte(unsigned long offset);
+unsigned short __readx18word(unsigned long offset);
+unsigned long __readx18dword(unsigned long offset);
+unsigned __int64 __readx18qword(unsigned long offset);
 #endif
 
 /*----------------------------------------------------------------------------*\
 |* Privileged intrinsics
 \*----------------------------------------------------------------------------*/
 #if defined(__i386__) || defined(__x86_64__)
 static __inline__ unsigned __int64 __DEFAULT_FN_ATTRS
 __readmsr(unsigned long __register) {

clang/test/CodeGen/arm64-microsoft-intrinsics.c

unsigned __int64 check__getReg(void) {
   unsigned volatile __int64 reg;
   reg = __getReg(18);
   reg = __getReg(31);
   return reg;
 }
 
 // CHECK-MSVC: call i64 @llvm.read_register.i64(metadata ![[MD2:.*]])
 // CHECK-MSVC: call i64 @llvm.read_register.i64(metadata ![[MD3:.*]])
+
+unsigned char check__readx18byte(unsigned long offset) {
+  return __readx18byte(offset);
+}
+
+// CHECK-MSVC: %[[OFFSET_ADDR:.*]] = alloca i32, align 4
+// CHECK-MSVC: store i32 %offset, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[X18:.*]] = call i64 @llvm.read_register.i64(metadata ![[MD2]])
+// CHECK-MSVC: %[[X18_AS_PTR:.*]] = inttoptr i64 %[[X18]] to i8*
+// CHECK-MSVC: %[[OFFSET:.*]] = load i32, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[ZEXT_OFFSET:.*]] = zext i32 %[[OFFSET]] to i64
+// CHECK-MSVC: %[[PTR:.*]] = getelementptr i8, i8* %[[X18_AS_PTR]], i64 %[[ZEXT_OFFSET]]
+// CHECK-MSVC: %[[RETVAL:.*]] = load i8, i8* %[[PTR]], align 1
+// CHECK-MSVC: ret i8 %[[RETVAL]]
+
+unsigned short check__readx18word(unsigned long offset) {
+  return __readx18word(offset);
+}
+
+// CHECK-MSVC: %[[OFFSET_ADDR:.*]] = alloca i32, align 4
+// CHECK-MSVC: store i32 %offset, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[X18:.*]] = call i64 @llvm.read_register.i64(metadata ![[MD2]])
+// CHECK-MSVC: %[[X18_AS_PTR:.*]] = inttoptr i64 %[[X18]] to i8*
+// CHECK-MSVC: %[[OFFSET:.*]] = load i32, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[ZEXT_OFFSET:.*]] = zext i32 %[[OFFSET]] to i64
+// CHECK-MSVC: %[[PTR:.*]] = getelementptr i8, i8* %[[X18_AS_PTR]], i64 %[[ZEXT_OFFSET]]
+// CHECK-MSVC: %[[BITCAST_PTR:.*]] = bitcast i8* %[[PTR]] to i16*
+// CHECK-MSVC: %[[RETVAL:.*]] = load i16, i16* %[[BITCAST_PTR]], align 2
+// CHECK-MSVC: ret i16 %[[RETVAL]]
+
+unsigned long check__readx18dword(unsigned long offset) {
+  return __readx18dword(offset);
+}
+
+// CHECK-MSVC: %[[OFFSET_ADDR:.*]] = alloca i32, align 4
+// CHECK-MSVC: store i32 %offset, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[X18:.*]] = call i64 @llvm.read_register.i64(metadata ![[MD2]])
+// CHECK-MSVC: %[[X18_AS_PTR:.*]] = inttoptr i64 %[[X18]] to i8*
+// CHECK-MSVC: %[[OFFSET:.*]] = load i32, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[ZEXT_OFFSET:.*]] = zext i32 %[[OFFSET]] to i64
+// CHECK-MSVC: %[[PTR:.*]] = getelementptr i8, i8* %[[X18_AS_PTR]], i64 %[[ZEXT_OFFSET]]
+// CHECK-MSVC: %[[BITCAST_PTR:.*]] = bitcast i8* %[[PTR]] to i32*
+// CHECK-MSVC: %[[RETVAL:.*]] = load i32, i32* %[[BITCAST_PTR]], align 4
+// CHECK-MSVC: ret i32 %[[RETVAL]]
+
+unsigned __int64 check__readx18qword(unsigned long offset) {
+  return __readx18qword(offset);
+}
+
+// CHECK-MSVC: %[[OFFSET_ADDR:.*]] = alloca i32, align 4
+// CHECK-MSVC: store i32 %offset, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[X18:.*]] = call i64 @llvm.read_register.i64(metadata ![[MD2]])
+// CHECK-MSVC: %[[X18_AS_PTR:.*]] = inttoptr i64 %[[X18]] to i8*
+// CHECK-MSVC: %[[OFFSET:.*]] = load i32, i32* %[[OFFSET_ADDR]], align 4
+// CHECK-MSVC: %[[ZEXT_OFFSET:.*]] = zext i32 %[[OFFSET]] to i64
+// CHECK-MSVC: %[[PTR:.*]] = getelementptr i8, i8* %[[X18_AS_PTR]], i64 %[[ZEXT_OFFSET]]
+// CHECK-MSVC: %[[BITCAST_PTR:.*]] = bitcast i8* %[[PTR]] to i64*
+// CHECK-MSVC: %[[RETVAL:.*]] = load i64, i64* %[[BITCAST_PTR]], align 8
+// CHECK-MSVC: ret i64 %[[RETVAL]]
+
 // CHECK-MSVC: ![[MD2]] = !{!"x18"}
 // CHECK-MSVC: ![[MD3]] = !{!"sp"}