[CIR] Add support for __int128 type (#980)

This PR adds initial support for the `__int128` type. The `!cir.int`
type is extended to support 128-bit integer types.

This PR comes with a simple test that verifies the CIRGen and LLVM
lowering of `!s128i` and `!u128i` work.

Resolve #953 .

Author: Lancern

clang/include/clang/CIR/Dialect/IR/CIRTypes.td

@@ -38,9 +38,11 @@ def CIR_IntType : CIR_Type<"Int", "int",
     [DeclareTypeInterfaceMethods<DataLayoutTypeInterface>]> {
   let summary = "Integer type with arbitrary precision up to a fixed limit";
   let description = [{
-    CIR type that represents C/C++ primitive integer types.
-    Said types are: `char`, `short`, `int`, `long`, `long long`, and their \
-    unsigned variations.
+    CIR type that represents integer types with arbitrary precision.
+
+    Those integer types that are directly available in C/C++ standard are called
+    primitive integer types. Said types are: `signed char`, `short`, `int`,
+    `long`, `long long`, and their unsigned variations.
   }];
   let parameters = (ins "unsigned":$width, "bool":$isSigned);
   let hasCustomAssemblyFormat = 1;
@@ -62,7 +64,7 @@ def CIR_IntType : CIR_Type<"Int", "int",
     /// Returns a minimum bitwidth of cir::IntType
     static unsigned minBitwidth() { return 1; }
     /// Returns a maximum bitwidth of cir::IntType
-    static unsigned maxBitwidth() { return 64; }
+    static unsigned maxBitwidth() { return 128; }
 
     /// Returns true if cir::IntType that represents a primitive integer type
     /// can be constructed from the provided bitwidth.

clang/lib/CIR/CodeGen/CIRGenModule.cpp

@@ -120,6 +120,8 @@ CIRGenModule::CIRGenModule(mlir::MLIRContext &context,
       ::mlir::cir::IntType::get(builder.getContext(), 32, /*isSigned=*/true);
   SInt64Ty =
       ::mlir::cir::IntType::get(builder.getContext(), 64, /*isSigned=*/true);
+  SInt128Ty =
+      ::mlir::cir::IntType::get(builder.getContext(), 128, /*isSigned=*/true);
 
   // Initialize CIR unsigned integer types cache.
   UInt8Ty =
@@ -130,6 +132,8 @@ CIRGenModule::CIRGenModule(mlir::MLIRContext &context,
       ::mlir::cir::IntType::get(builder.getContext(), 32, /*isSigned=*/false);
   UInt64Ty =
       ::mlir::cir::IntType::get(builder.getContext(), 64, /*isSigned=*/false);
+  UInt128Ty =
+      ::mlir::cir::IntType::get(builder.getContext(), 128, /*isSigned=*/false);
 
   VoidTy = ::mlir::cir::VoidType::get(builder.getContext());
 

clang/lib/CIR/CodeGen/CIRGenTypeCache.h

@@ -31,10 +31,10 @@ struct CIRGenTypeCache {
 
   /// void
   mlir::cir::VoidType VoidTy;
-  // char, int, short, long
-  mlir::cir::IntType SInt8Ty, SInt16Ty, SInt32Ty, SInt64Ty;
-  // usigned char, unsigned, unsigned short, unsigned long
-  mlir::cir::IntType UInt8Ty, UInt16Ty, UInt32Ty, UInt64Ty;
+  // char, int, short, long, __int128
+  mlir::cir::IntType SInt8Ty, SInt16Ty, SInt32Ty, SInt64Ty, SInt128Ty;
+  // usigned char, unsigned, unsigned short, unsigned long, unsigned __int128
+  mlir::cir::IntType UInt8Ty, UInt16Ty, UInt32Ty, UInt64Ty, UInt128Ty;
   /// half, bfloat, float, double, fp80
   mlir::cir::FP16Type FP16Ty;
   mlir::cir::BF16Type BFloat16Ty;

clang/lib/CIR/CodeGen/CIRGenTypes.cpp

@@ -503,9 +503,10 @@ mlir::Type CIRGenTypes::ConvertType(QualType T) {
       break;
 
     case BuiltinType::UInt128:
+      ResultType = CGM.UInt128Ty;
+      break;
     case BuiltinType::Int128:
-      assert(0 && "not implemented");
-      // FIXME: ResultType = Builder.getIntegerType(128);
+      ResultType = CGM.SInt128Ty;
       break;
 
 #define IMAGE_TYPE(ImgType, Id, SingletonId, Access, Suffix)                   \

clang/lib/CIR/Dialect/IR/CIRTypes.cpp

@@ -604,8 +604,9 @@ Type IntType::parse(mlir::AsmParser &parser) {
   // Fetch integer size.
   if (parser.parseInteger(width))
     return {};
-  if (width < 1 || width > 64) {
-    parser.emitError(loc, "expected integer width to be from 1 up to 64");
+  if (width < IntType::minBitwidth() || width > IntType::maxBitwidth()) {
+    parser.emitError(loc, "expected integer width to be from ")
+        << IntType::minBitwidth() << " up to " << IntType::maxBitwidth();
     return {};
   }
 
@@ -643,7 +644,8 @@ IntType::verify(llvm::function_ref<mlir::InFlightDiagnostic()> emitError,
 
   if (width < IntType::minBitwidth() || width > IntType::maxBitwidth()) {
     emitError() << "IntType only supports widths from "
-                << IntType::minBitwidth() << "up to " << IntType::maxBitwidth();
+                << IntType::minBitwidth() << " up to "
+                << IntType::maxBitwidth();
     return mlir::failure();
   }
 

clang/test/CIR/CodeGen/int128.cpp

@@ -0,0 +1,74 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm -fno-clangir-call-conv-lowering %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll --check-prefix=LLVM %s
+
+// TODO: remove the -fno-clangir-call-conv-lowering flag when ABI lowering for
+//       int128 is supported.
+
+// CHECK-LABEL: @_Z5test1n
+// LLVM-LABEL: @_Z5test1n
+__int128 test1(__int128 x) {
+  return x;
+  // CHECK: cir.return %{{.+}} : !s128i
+  // LLVM: ret i128 %{{.+}}
+}
+
+// CHECK-LABEL: @_Z5test2o
+// LLVM-LABEL: @_Z5test2o
+unsigned __int128 test2(unsigned __int128 x) {
+  return x;
+  // CHECK: cir.return %{{.+}} : !u128i
+  // LLVM: ret i128 %{{.+}}
+}
+
+// CHECK-LABEL: @_Z11unary_arithn
+// LLVM-LABEL: @_Z11unary_arithn
+__int128 unary_arith(__int128 x) {
+  return ++x;
+  // CHECK: %{{.+}} = cir.unary(inc, %{{.+}}) : !s128i, !s128i
+  // LLVM: %{{.+}} = add i128 %{{.+}}, 1
+}
+
+// CHECK-LABEL: @_Z12binary_arithnn
+// LLVM-LABEL: @_Z12binary_arithnn
+__int128 binary_arith(__int128 x, __int128 y) {
+  return x + y;
+  // CHECK: %{{.+}} = cir.binop(add, %{{.+}}, %{{.+}}) nsw : !s128i
+  // LLVM: %{{.+}} = add nsw i128 %{{.+}}, %{{.+}}
+}
+
+volatile int int_var;
+volatile double double_var;
+
+// CHECK-LABEL: @_Z19integral_conversionn
+// LLVM-LABEL: @_Z19integral_conversionn
+__int128 integral_conversion(__int128 x) {
+  int_var = x;
+  // CHECK: %[[#VAL:]] = cir.cast(integral, %{{.+}} : !s128i), !s32i
+  // LLVM: %{{.+}} = trunc i128 %{{.+}} to i32
+
+  return int_var;
+  // CHECK: %{{.+}} = cir.cast(integral, %{{.+}} : !s32i), !s128i
+  // LLVM: %{{.+}} = sext i32 %{{.+}} to i128
+}
+
+// CHECK-LABEL: @_Z16float_conversionn
+// LLVM-LABEL: @_Z16float_conversionn
+__int128 float_conversion(__int128 x) {
+  double_var = x;
+  // CHECK: %[[#VAL:]] = cir.cast(int_to_float, %{{.+}} : !s128i), !cir.double
+  // LLVM: %{{.+}} = sitofp i128 %{{.+}} to double
+
+  return double_var;
+  // CHECK: %{{.+}} = cir.cast(float_to_int, %{{.+}} : !cir.double), !s128i
+  // LLVM: %{{.+}} = fptosi double %{{.+}} to i128
+}
+
+// CHECK-LABEL: @_Z18boolean_conversionn
+// LLVM-LABEL: @_Z18boolean_conversionn
+bool boolean_conversion(__int128 x) {
+  return x;
+  // CHECK: %{{.+}} = cir.cast(int_to_bool, %{{.+}} : !s128i), !cir.bool
+  // LLVM: %{{.+}} = icmp ne i128 %{{.+}}, 0
+}

clang/test/CIR/IR/invalid.cir

@@ -593,8 +593,8 @@ module {
 // // -----
 
 module {
-  // expected-error@below {{expected integer width to be from 1 up to 64}}
-  cir.func @l0(%arg0: !cir.int<s, 128>) -> () {
+  // expected-error@below {{expected integer width to be from 1 up to 128}}
+  cir.func @l0(%arg0: !cir.int<s, 256>) -> () {
     cir.return
   }
 }