Introduce the asmjs-unknown-emscripten target triple.

Notable changes from le32-unknown-nacl so far include:
 - Set i32 as the legal integer set, to help the optimizer avoid creating
   needlessly inefficient code for asm.js.
 - We can use llvm.pow.
 - Don't predefine __ELF__ or __pnacl__ so that we don't need to undefine
   them later.
 - Do predefine asm.js and Emscripten macros, so that we don't need to
   define them later.
 - Don't provide __has_feature(pnacl).

8 files changed, 409 insertions, 4 deletions

diff --git a/lib/Basic/Targets.cpp b/lib/Basic/Targets.cpp
index a622a11aa5..43a21f2734 100644
--- a/lib/Basic/Targets.cpp
+++ b/lib/Basic/Targets.cpp
@@ -239,6 +239,39 @@ public:
   }
 };
 
+// Emscripten target
+template<typename Target>
+class EmscriptenTargetInfo : public OSTargetInfo<Target> {
+protected:
+  virtual void getOSDefines(const LangOptions &Opts, const llvm::Triple &Triple,
+                            MacroBuilder &Builder) const {
+    // A macro for the platform.
+    Builder.defineMacro("__EMSCRIPTEN__");
+    // Earlier versions of Emscripten defined this, so we continue to define it
+    // for compatibility, for now. Users should ideally prefer __EMSCRIPTEN__.
+    Builder.defineMacro("EMSCRIPTEN");
+    // A common platform macro.
+    if (Opts.POSIXThreads)
+      Builder.defineMacro("_REENTRANT");
+    // Follow g++ convention and predefine _GNU_SOURCE for C++.
+    if (Opts.CPlusPlus)
+      Builder.defineMacro("_GNU_SOURCE");
+
+    // Emscripten's software environment and the asm.js runtime aren't really
+    // Unix per se, but they're perhaps more Unix-like than what software
+    // expects when "unix" is *not* defined.
+    DefineStd(Builder, "unix", Opts);
+  }
+public:
+  explicit EmscriptenTargetInfo(const std::string& triple)
+    : OSTargetInfo<Target>(triple) {
+    // Emcripten currently does prepend a prefix to user labels, but this is
+    // handled outside of clang. TODO: Handling this within clang may be
+    // beneficial.
+    this->UserLabelPrefix = "";
+  }
+};
+
 // FreeBSD Target
 template<typename Target>
 class FreeBSDTargetInfo : public OSTargetInfo<Target> {
@@ -5042,6 +5075,82 @@ public:
 } // end anonymous namespace.
 
 namespace {
+class AsmJSTargetInfo : public TargetInfo {
+public:
+  explicit AsmJSTargetInfo(const std::string& triple) : TargetInfo(triple) {
+    BigEndian = false;
+    this->LongAlign = 32;
+    this->LongWidth = 32;
+    this->PointerAlign = 32;
+    this->PointerWidth = 32;
+    this->IntMaxType = TargetInfo::SignedLongLong;
+    this->UIntMaxType = TargetInfo::UnsignedLongLong;
+    this->Int64Type = TargetInfo::SignedLongLong;
+    this->DoubleAlign = 64;
+    this->LongDoubleWidth = 64;
+    this->LongDoubleAlign = 64;
+    this->SizeType = TargetInfo::UnsignedInt;
+    this->PtrDiffType = TargetInfo::SignedInt;
+    this->IntPtrType = TargetInfo::SignedInt;
+    this->RegParmMax = 0; // Disallow regparm
+
+    // Set the native integer widths set to just i32, since that's currently
+    // the only integer type we can do arithmetic on without masking or
+    // splitting.
+    //
+    // Set the required alignment for 128-bit vectors to just 4 bytes, based on
+    // the direction suggested here:
+    //   https://bugzilla.mozilla.org/show_bug.cgi?id=904913#c21
+    // We can still set the preferred alignment to 16 bytes though.
+    DescriptionString = "e-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-"
+                        "f32:32:32-f64:64:64-p:32:32:32-v128:32:128-n32";
+  }
+
+  void getDefaultFeatures(llvm::StringMap<bool> &Features) const {
+  }
+  virtual void getArchDefines(const LangOptions &Opts,
+                              MacroBuilder &Builder) const {
+    Builder.defineMacro("__asmjs__");
+  }
+  virtual void getTargetDefines(const LangOptions &Opts,
+                                MacroBuilder &Builder) const {
+    Builder.defineMacro("__LITTLE_ENDIAN__");
+    getArchDefines(Opts, Builder);
+  }
+  virtual void getTargetBuiltins(const Builtin::Info *&Records,
+                                 unsigned &NumRecords) const {
+  }
+  virtual BuiltinVaListKind getBuiltinVaListKind() const {
+    // Reuse PNaCl's va_list lowering.
+    return TargetInfo::PNaClABIBuiltinVaList;
+  }
+  virtual void getGCCRegNames(const char * const *&Names,
+                              unsigned &NumNames) const {
+    Names = NULL;
+    NumNames = 0;
+  }
+  virtual void getGCCRegAliases(const GCCRegAlias *&Aliases,
+                                unsigned &NumAliases) const {
+    Aliases = NULL;
+    NumAliases = 0;
+  }
+  virtual bool validateAsmConstraint(const char *&Name,
+                                     TargetInfo::ConstraintInfo &Info) const {
+    return false;
+  }
+  virtual const char *getClobbers() const {
+    return "";
+  }
+  virtual bool isCLZForZeroUndef() const {
+    // Today we do clz in software, so we just do the right thing. With ES6,
+    // we'll get Math.clz32, which is to be defined to do the right thing:
+    // http://esdiscuss.org/topic/rename-number-prototype-clz-to-math-clz#content-36
+    return false;
+  }
+};
+} // end anonymous namespace.
+
+namespace {
 class PNaClTargetInfo : public TargetInfo {
 public:
   PNaClTargetInfo(const std::string& triple) : TargetInfo(triple) {
@@ -5315,6 +5424,14 @@ static TargetInfo *AllocateTarget(const std::string &T) {
       return new Mips64ELTargetInfo(T);
     }
 
+  case llvm::Triple::asmjs:
+    switch (os) {
+      case llvm::Triple::Emscripten:
+        return new EmscriptenTargetInfo<AsmJSTargetInfo>(T);
+      default:
+        return NULL;
+    }
+
   case llvm::Triple::le32:
     switch (os) {
       case llvm::Triple::NaCl:
diff --git a/lib/CodeGen/ItaniumCXXABI.cpp b/lib/CodeGen/ItaniumCXXABI.cpp
index 8a7f551e68..3767e4f278 100644
--- a/lib/CodeGen/ItaniumCXXABI.cpp
+++ b/lib/CodeGen/ItaniumCXXABI.cpp
@@ -232,15 +232,17 @@ CodeGen::CGCXXABI *CodeGen::CreateItaniumCXXABI(CodeGenModule &CGM) {
                              /* UseARMGuardVarABI = */ true);
 
   case TargetCXXABI::GenericItanium:
-    if (CGM.getContext().getTargetInfo().getTriple().getArch()
-        == llvm::Triple::le32) {
-      // For PNaCl, use ARM-style method pointers so that PNaCl code
+    switch (CGM.getContext().getTargetInfo().getTriple().getArch()) {
+    case llvm::Triple::le32:
+    case llvm::Triple::asmjs:
+      // Use ARM-style method pointers so that generated code
       // does not assume anything about the alignment of function
       // pointers.
       return new ItaniumCXXABI(CGM, /* UseARMMethodPtrABI = */ true,
                                /* UseARMGuardVarABI = */ false);
+    default:
+      return new ItaniumCXXABI(CGM);
     }
-    return new ItaniumCXXABI(CGM);
 
   case TargetCXXABI::Microsoft:
     llvm_unreachable("Microsoft ABI is not Itanium-based");
diff --git a/lib/CodeGen/TargetInfo.cpp b/lib/CodeGen/TargetInfo.cpp
index 4fb888a767..5aa982e255 100644
--- a/lib/CodeGen/TargetInfo.cpp
+++ b/lib/CodeGen/TargetInfo.cpp
@@ -411,6 +411,62 @@ ABIArgInfo DefaultABIInfo::classifyReturnType(QualType RetTy) const {
 }
 
 //===----------------------------------------------------------------------===//
+// Emscripten ABI Implementation
+//
+// This is a very simple ABI that relies a lot on DefaultABIInfo.
+//===----------------------------------------------------------------------===//
+
+class EmscriptenABIInfo : public DefaultABIInfo {
+ public:
+  explicit EmscriptenABIInfo(CodeGen::CodeGenTypes &CGT) : DefaultABIInfo(CGT) {}
+
+  ABIArgInfo classifyReturnType(QualType RetTy) const;
+  ABIArgInfo classifyArgumentType(QualType RetTy) const;
+};
+
+class EmscriptenTargetCodeGenInfo : public TargetCodeGenInfo {
+ public:
+  explicit EmscriptenTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
+    : TargetCodeGenInfo(new EmscriptenABIInfo(CGT)) {}
+
+  // TODO: Re-evaluate whether these hacks, borrowed from PNaCl, are necessary.
+  bool addAsmMemoryAroundSyncSynchronize() const { return true; }
+  bool asmMemoryIsFence() const { return true; }
+};
+
+/// \brief Classify argument of given type \p Ty.
+ABIArgInfo EmscriptenABIInfo::classifyArgumentType(QualType Ty) const {
+  if (isAggregateTypeForABI(Ty)) {
+    if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, CGT))
+      return ABIArgInfo::getIndirect(0, RAA == CGCXXABI::RAA_DirectInMemory);
+    return ABIArgInfo::getIndirect(0);
+  }
+
+  // We can handle floating-point values directly.
+  if (Ty->isFloatingType())
+    return ABIArgInfo::getDirect();
+
+  // Otherwise just do the default thing.
+  return DefaultABIInfo::classifyArgumentType(Ty);
+}
+
+ABIArgInfo EmscriptenABIInfo::classifyReturnType(QualType RetTy) const {
+  if (isAggregateTypeForABI(RetTy)) {
+    // As an optimization, lower single-element structs to just return a
+    // regular value.
+    if (const Type *SeltTy = isSingleElementStruct(RetTy, getContext()))
+      return ABIArgInfo::getDirect(CGT.ConvertType(QualType(SeltTy, 0)));
+  }
+
+  // We can handle floating-point values directly.
+  if (RetTy->isFloatingType())
+    return ABIArgInfo::getDirect();
+
+  // Otherwise just do the default thing.
+  return DefaultABIInfo::classifyReturnType(RetTy);
+}
+
+//===----------------------------------------------------------------------===//
 // le32/PNaCl bitcode ABI Implementation
 //
 // This is a simplified version of the x86_32 ABI.  Arguments and return values
@@ -5101,6 +5157,9 @@ const TargetCodeGenInfo &CodeGenModule::getTargetCodeGenInfo() {
   default:
     return *(TheTargetCodeGenInfo = new DefaultTargetCodeGenInfo(Types));
 
+  case llvm::Triple::asmjs:
+    return *(TheTargetCodeGenInfo = new EmscriptenTargetCodeGenInfo(Types));
+
   case llvm::Triple::le32:
     return *(TheTargetCodeGenInfo = new PNaClTargetCodeGenInfo(Types));
   case llvm::Triple::mips:
diff --git a/test/CodeGen/emscripten-arguments.c b/test/CodeGen/emscripten-arguments.c
new file mode 100644
index 0000000000..afa0d9c57b
--- /dev/null
+++ b/test/CodeGen/emscripten-arguments.c
@@ -0,0 +1,61 @@
+// RUN: %clang_cc1 -triple asmjs-unknown-emscripten %s -emit-llvm -o - | FileCheck %s
+
+// Basic argument/attribute tests for asmjs/Emscripten
+
+// CHECK: define void @f0(i32 %i, i32 %j, double %k)
+void f0(int i, long j, double k) {}
+
+typedef struct {
+  int aa;
+  int bb;
+} s1;
+// Structs should be passed byval and not split up
+// CHECK: define void @f1(%struct.s1* byval %i)
+void f1(s1 i) {}
+
+typedef struct {
+  int cc;
+} s2;
+// Structs should be returned sret and not simplified by the frontend
+// CHECK: define void @f2(%struct.s2* noalias sret %agg.result)
+s2 f2() {
+  s2 foo;
+  return foo;
+}
+
+// CHECK: define void @f3(i64 %i)
+void f3(long long i) {}
+
+// i8/i16 should be signext, i32 and higher should not
+// CHECK: define void @f4(i8 signext %a, i16 signext %b)
+void f4(char a, short b) {}
+
+// CHECK: define void @f5(i8 zeroext %a, i16 zeroext %b)
+void f5(unsigned char a, unsigned short b) {}
+
+
+enum my_enum {
+  ENUM1,
+  ENUM2,
+  ENUM3,
+};
+// Enums should be treated as the underlying i32
+// CHECK: define void @f6(i32 %a)
+void f6(enum my_enum a) {}
+
+union simple_union {
+  int a;
+  char b;
+};
+// Unions should be passed as byval structs
+// CHECK: define void @f7(%union.simple_union* byval %s)
+void f7(union simple_union s) {}
+
+typedef struct {
+  int b4 : 4;
+  int b3 : 3;
+  int b8 : 8;
+} bitfield1;
+// Bitfields should be passed as byval structs
+// CHECK: define void @f8(%struct.bitfield1* byval %bf1)
+void f8(bitfield1 bf1) {}
diff --git a/test/CodeGen/emscripten-regparm.c b/test/CodeGen/emscripten-regparm.c
new file mode 100644
index 0000000000..20ffc9dec0
--- /dev/null
+++ b/test/CodeGen/emscripten-regparm.c
@@ -0,0 +1,4 @@
+// RUN: %clang_cc1 -triple asmjs-unknown-emscripten %s -fsyntax-only -verify
+
+void __attribute__((regparm(2))) fc_f1(int i, int j, int k) {} // expected-error{{'regparm' is not valid on this platform}}
+
diff --git a/test/CodeGenCXX/member-function-pointers.cpp b/test/CodeGenCXX/member-function-pointers.cpp
index 23a7b0009f..c29f0e75e5 100644
--- a/test/CodeGenCXX/member-function-pointers.cpp
+++ b/test/CodeGenCXX/member-function-pointers.cpp
@@ -6,6 +6,8 @@
 
 // PNaCl uses the same representation of method pointers as ARM.
 // RUN: %clang_cc1 %s -emit-llvm -o - -triple=le32-unknown-nacl | FileCheck -check-prefix GLOBAL-ARM %s
+// Emscripten uses the same representation of method pointers as ARM.
+// RUN: %clang_cc1 %s -emit-llvm -o - -triple=asmjs-unknown-emscripten | FileCheck -check-prefix GLOBAL-ARM %s
 
 struct A { int a; void f(); virtual void vf1(); virtual void vf2(); };
 struct B { int b; virtual void g(); };
diff --git a/test/CodeGenCXX/static-init-emscripten.cpp b/test/CodeGenCXX/static-init-emscripten.cpp
new file mode 100644
index 0000000000..b9d0271ae1
--- /dev/null
+++ b/test/CodeGenCXX/static-init-emscripten.cpp
@@ -0,0 +1,14 @@
+// RUN: %clang_cc1 -emit-llvm -triple=asmjs-unknown-emscripten -o - %s | FileCheck %s
+
+int f();
+
+// Test that Emscripten uses the Itanium/x86 ABI in which the static
+// variable's guard variable is tested via "load i8 and compare with
+// zero" rather than the ARM ABI which uses "load i32 and test the
+// bottom bit".
+void g() {
+  static int a = f();
+}
+// CHECK: load atomic i8* bitcast (i64* @_ZGVZ1gvE1a to i8*) acquire
+// CHECK-NEXT: %guard.uninitialized = icmp eq i8 %0, 0
+// CHECK-NEXT: br i1 %guard.uninitialized, label %init.check, label %init.end
diff --git a/test/Driver/asmjs-unknown-emscripten.cpp b/test/Driver/asmjs-unknown-emscripten.cpp
new file mode 100644
index 0000000000..fb94f0ecf9
--- /dev/null
+++ b/test/Driver/asmjs-unknown-emscripten.cpp
@@ -0,0 +1,146 @@
+// RUN: %clang -target asmjs-unknown-emscripten -ccc-echo %s -emit-llvm-only -c 2>&1 | FileCheck %s -check-prefix=ECHO
+// RUN: %clang -target asmjs-unknown-emscripten %s -emit-llvm -S -c -o - | FileCheck %s
+// RUN: %clang -target asmjs-unknown-emscripten %s -emit-llvm -S -c -pthread -o - | FileCheck %s -check-prefix=THREADS
+
+// ECHO: {{.*}} -cc1 {{.*}}asmjs-unknown-emscripten.c
+
+// Check platform defines
+#include <stdarg.h>
+#include <stddef.h>
+
+extern "C" {
+
+// CHECK: @align_c = global i32 1
+int align_c = __alignof(char);
+
+// CHECK: @align_s = global i32 2
+int align_s = __alignof(short);
+
+// CHECK: @align_i = global i32 4
+int align_i = __alignof(int);
+
+// CHECK: @align_l = global i32 4
+int align_l = __alignof(long);
+
+// CHECK: @align_ll = global i32 8
+int align_ll = __alignof(long long);
+
+// CHECK: @align_p = global i32 4
+int align_p = __alignof(void*);
+
+// CHECK: @align_f = global i32 4
+int align_f = __alignof(float);
+
+// CHECK: @align_d = global i32 8
+int align_d = __alignof(double);
+
+// CHECK: @align_ld = global i32 8
+int align_ld = __alignof(long double);
+
+// CHECK: @align_vl = global i32 4
+int align_vl = __alignof(va_list);
+
+// CHECK: __LITTLE_ENDIAN__defined
+#ifdef __LITTLE_ENDIAN__
+void __LITTLE_ENDIAN__defined() {}
+#endif
+
+// CHECK: __native_client__defined
+#ifdef __native_client__
+void __native_client__defined() {}
+#endif
+
+// CHECK: __le32__defined
+#ifdef __le32__
+void __le32__defined() {}
+#endif
+
+// CHECK: __pnacl__defined
+#ifdef __pnacl__
+void __pnacl__defined() {}
+#endif
+
+// CHECK: unixdefined
+#ifdef unix
+void unixdefined() {}
+#endif
+
+// CHECK: __ELF__defined
+#ifdef __ELF__
+void __ELF__defined() {}
+#endif
+
+// CHECK: _GNU_SOURCEdefined
+#ifdef _GNU_SOURCE
+void _GNU_SOURCEdefined() {}
+#endif
+
+// THREADS: _REENTRANTdefined
+// CHECK: _REENTRANTundefined
+#ifdef _REENTRANT
+void _REENTRANTdefined() {}
+#else
+void _REENTRANTundefined() {}
+#endif
+
+// Check types
+
+// CHECK: signext i8 @check_char()
+char check_char() { return 0; }
+
+// CHECK: signext i16 @check_short()
+short check_short() { return 0; }
+
+// CHECK: i32 @check_int()
+int check_int() { return 0; }
+
+// CHECK: i32 @check_long()
+long check_long() { return 0; }
+
+// CHECK: i64 @check_longlong()
+long long check_longlong() { return 0; }
+
+// CHECK: zeroext i8 @check_uchar()
+unsigned char check_uchar() { return 0; }
+
+// CHECK: zeroext i16 @check_ushort()
+unsigned short check_ushort() { return 0; }
+
+// CHECK: i32 @check_uint()
+unsigned int check_uint() { return 0; }
+
+// CHECK: i32 @check_ulong()
+unsigned long check_ulong() { return 0; }
+
+// CHECK: i64 @check_ulonglong()
+unsigned long long check_ulonglong() { return 0; }
+
+// CHECK: i32 @check_size_t()
+size_t check_size_t() { return 0; }
+
+// CHECK: float @check_float()
+float check_float() { return 0; }
+
+// CHECK: double @check_double()
+double check_double() { return 0; }
+
+// CHECK: double @check_longdouble()
+long double check_longdouble() { return 0; }
+
+}
+
+template<int> void Switch();
+template<> void Switch<4>();
+template<> void Switch<8>();
+template<> void Switch<16>();
+
+void check_pointer_size() {
+  // CHECK: SwitchILi4
+  Switch<sizeof(void*)>();
+
+  // CHECK: SwitchILi8
+  Switch<sizeof(long long)>();
+
+  // CHECK: SwitchILi16
+  Switch<sizeof(va_list)>();
+}