X64: Addition binary operation.

Review URL: http://codereview.chromium.org/146022


git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@2255 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/x64/assembler-x64-inl.h

@@ -29,6 +29,7 @@
 #define V8_X64_ASSEMBLER_X64_INL_H_
 
 #include "cpu.h"
+#include "memory.h"
 
 namespace v8 {
 namespace internal {
@@ -74,11 +75,21 @@ void Assembler::emit_rex_64(Register reg, Register rm_reg) {
 }
 
 
+void Assembler::emit_rex_64(XMMRegister reg, Register rm_reg) {
+  emit(0x48 | (reg.code() & 0x8) >> 1 | rm_reg.code() >> 3);
+}
+
+
 void Assembler::emit_rex_64(Register reg, const Operand& op) {
   emit(0x48 | reg.high_bit() << 2 | op.rex_);
 }
 
 
+void Assembler::emit_rex_64(XMMRegister reg, const Operand& op) {
+  emit(0x48 | (reg.code() & 0x8) >> 1 | op.rex_);
+}
+
+
 void Assembler::emit_rex_64(Register rm_reg) {
   ASSERT_EQ(rm_reg.code() & 0xf, rm_reg.code());
   emit(0x48 | rm_reg.high_bit());
@@ -122,6 +133,24 @@ void Assembler::emit_optional_rex_32(Register reg, const Operand& op) {
 }
 
 
+void Assembler::emit_optional_rex_32(XMMRegister reg, const Operand& op) {
+  byte rex_bits =  (reg.code() & 0x8) >> 1 | op.rex_;
+  if (rex_bits != 0) emit(0x40 | rex_bits);
+}
+
+
+void Assembler::emit_optional_rex_32(XMMRegister reg, XMMRegister base) {
+  byte rex_bits =  (reg.code() & 0x8) >> 1 | (base.code() & 0x8) >> 3;
+  if (rex_bits != 0) emit(0x40 | rex_bits);
+}
+
+
+void Assembler::emit_optional_rex_32(XMMRegister reg, Register base) {
+  byte rex_bits =  (reg.code() & 0x8) >> 1 | (base.code() & 0x8) >> 3;
+  if (rex_bits != 0) emit(0x40 | rex_bits);
+}
+
+
 void Assembler::emit_optional_rex_32(Register rm_reg) {
   if (rm_reg.high_bit()) emit(0x41);
 }

src/x64/assembler-x64.cc

@@ -113,20 +113,20 @@ Operand::Operand(Register base,
 }
 
 
-// Safe default is no features.
-// TODO(X64): Safe defaults include SSE2 for X64.
-uint64_t CpuFeatures::supported_ = 0;
+// The required user mode extensions in X64 are (from AMD64 ABI Table A.1):
+//   fpu, tsc, cx8, cmov, mmx, sse, sse2, fxsr, syscall
+uint64_t CpuFeatures::supported_ = kDefaultCpuFeatures;
 uint64_t CpuFeatures::enabled_ = 0;
 
 void CpuFeatures::Probe()  {
   ASSERT(Heap::HasBeenSetup());
-  ASSERT(supported_ == 0);
+  ASSERT(supported_ == kDefaultCpuFeatures);
   if (Serializer::enabled()) return;  // No features if we might serialize.
 
   Assembler assm(NULL, 0);
   Label cpuid, done;
 #define __ assm.
-  // Save old esp, since we are going to modify the stack.
+  // Save old rsp, since we are going to modify the stack.
   __ push(rbp);
   __ pushfq();
   __ push(rcx);
@@ -154,11 +154,11 @@ void CpuFeatures::Probe()  {
   // safe here.
   __ bind(&cpuid);
   __ movq(rax, Immediate(1));
-  supported_ = (1 << CPUID);
+  supported_ = kDefaultCpuFeatures | (1 << CPUID);
   { Scope fscope(CPUID);
     __ cpuid();
   }
-  supported_ = 0;
+  supported_ = kDefaultCpuFeatures;
 
   // Move the result from ecx:edx to rax and make sure to mark the
   // CPUID feature as supported.
@@ -187,6 +187,10 @@ void CpuFeatures::Probe()  {
   typedef uint64_t (*F0)();
   F0 probe = FUNCTION_CAST<F0>(Code::cast(code)->entry());
   supported_ = probe();
+  // SSE2 and CMOV must be available on an X64 CPU.
+  ASSERT(IsSupported(CPUID));
+  ASSERT(IsSupported(SSE2));
+  ASSERT(IsSupported(CMOV));
 }
 
 // -----------------------------------------------------------------------------
@@ -610,6 +614,57 @@ void Assembler::call(const Operand& op) {
 }
 
 
+void Assembler::cmovq(Condition cc, Register dst, Register src) {
+  // No need to check CpuInfo for CMOV support, it's a required part of the
+  // 64-bit architecture.
+  ASSERT(cc >= 0);  // Use mov for unconditional moves.
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  // Opcode: REX.W 0f 40 + cc /r
+  emit_rex_64(dst, src);
+  emit(0x0f);
+  emit(0x40 + cc);
+  emit_modrm(dst, src);
+}
+
+
+void Assembler::cmovq(Condition cc, Register dst, const Operand& src) {
+  ASSERT(cc >= 0);
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  // Opcode: REX.W 0f 40 + cc /r
+  emit_rex_64(dst, src);
+  emit(0x0f);
+  emit(0x40 + cc);
+  emit_operand(dst, src);
+}
+
+
+void Assembler::cmovl(Condition cc, Register dst, Register src) {
+  ASSERT(cc >= 0);
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  // Opcode: 0f 40 + cc /r
+  emit_optional_rex_32(dst, src);
+  emit(0x0f);
+  emit(0x40 + cc);
+  emit_modrm(dst, src);
+}
+
+
+void Assembler::cmovl(Condition cc, Register dst, const Operand& src) {
+  ASSERT(cc >= 0);
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  // Opcode: 0f 40 + cc /r
+  emit_optional_rex_32(dst, src);
+  emit(0x0f);
+  emit(0x40 + cc);
+  emit_operand(dst, src);
+}
+
+
+
 void Assembler::cpuid() {
   ASSERT(CpuFeatures::IsEnabled(CpuFeatures::CPUID));
   EnsureSpace ensure_space(this);
@@ -1752,6 +1807,15 @@ void Assembler::fnclex() {
 }
 
 
+void Assembler::sahf() {
+  // TODO(X64): Test for presence. Not all 64-bit intel CPU's have sahf
+  // in 64-bit mode. Test CpuID.
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0x9E);
+}
+
+
 void Assembler::emit_farith(int b1, int b2, int i) {
   ASSERT(is_uint8(b1) && is_uint8(b2));  // wrong opcode
   ASSERT(is_uint3(i));  // illegal stack offset
@@ -1759,6 +1823,166 @@ void Assembler::emit_farith(int b1, int b2, int i) {
   emit(b2 + i);
 }
 
+// SSE 2 operations
+
+void Assembler::movsd(const Operand& dst, XMMRegister src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);  // double
+  emit_optional_rex_32(src, dst);
+  emit(0x0F);
+  emit(0x11);  // store
+  emit_sse_operand(src, dst);
+}
+
+
+void Assembler::movsd(Register dst, XMMRegister src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);  // double
+  emit_optional_rex_32(src, dst);
+  emit(0x0F);
+  emit(0x11);  // store
+  emit_sse_operand(src, dst);
+}
+
+
+void Assembler::movsd(XMMRegister dst, Register src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);  // double
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x10);  // load
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::movsd(XMMRegister dst, const Operand& src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);  // double
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x10);  // load
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::cvttss2si(Register dst, const Operand& src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF3);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x2C);
+  emit_operand(dst, src);
+}
+
+
+void Assembler::cvttsd2si(Register dst, const Operand& src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x2C);
+  emit_operand(dst, src);
+}
+
+
+void Assembler::cvtlsi2sd(XMMRegister dst, const Operand& src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x2A);
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::cvtlsi2sd(XMMRegister dst, Register src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x2A);
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::cvtqsi2sd(XMMRegister dst, Register src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_rex_64(dst, src);
+  emit(0x0F);
+  emit(0x2A);
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::addsd(XMMRegister dst, XMMRegister src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x58);
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::mulsd(XMMRegister dst, XMMRegister src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x59);
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::subsd(XMMRegister dst, XMMRegister src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x5C);
+  emit_sse_operand(dst, src);
+}
+
+
+void Assembler::divsd(XMMRegister dst, XMMRegister src) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit(0xF2);
+  emit_optional_rex_32(dst, src);
+  emit(0x0F);
+  emit(0x5E);
+  emit_sse_operand(dst, src);
+}
+
+
+
+void Assembler::emit_sse_operand(XMMRegister reg, const Operand& adr) {
+  Register ireg = { reg.code() };
+  emit_operand(ireg, adr);
+}
+
+
+void Assembler::emit_sse_operand(XMMRegister dst, XMMRegister src) {
+  emit(0xC0 | (dst.code() << 3) | src.code());
+}
+
+void Assembler::emit_sse_operand(XMMRegister dst, Register src) {
+  emit(0xC0 | (dst.code() << 3) | src.code());
+}
+
 
 // Relocation information implementations
 

src/x64/assembler-x64.h

@@ -156,7 +156,7 @@ extern MMXRegister mm15;
 
 
 struct XMMRegister {
-  bool is_valid() const  { return 0 <= code_ && code_ < 2; }
+  bool is_valid() const  { return 0 <= code_ && code_ < 16; }
   int code() const  {
     ASSERT(is_valid());
     return code_;
@@ -330,11 +330,11 @@ class Operand BASE_EMBEDDED {
 // CpuFeatures keeps track of which features are supported by the target CPU.
 // Supported features must be enabled by a Scope before use.
 // Example:
-//   if (CpuFeatures::IsSupported(SSE2)) {
-//     CpuFeatures::Scope fscope(SSE2);
-//     // Generate SSE2 floating point code.
+//   if (CpuFeatures::IsSupported(SSE3)) {
+//     CpuFeatures::Scope fscope(SSE3);
+//     // Generate SSE3 floating point code.
 //   } else {
-//     // Generate standard x87 floating point code.
+//     // Generate standard x87 or SSE2 floating point code.
 //   }
 class CpuFeatures : public AllStatic {
  public:
@@ -371,6 +371,10 @@ class CpuFeatures : public AllStatic {
 #endif
   };
  private:
+  // Safe defaults include SSE2 and CMOV for X64. It is always available, if
+  // anyone checks, but they shouldn't need to check.
+  static const uint64_t kDefaultCpuFeatures =
+      (1 << CpuFeatures::SSE2 | 1 << CpuFeatures::CMOV);
   static uint64_t supported_;
   static uint64_t enabled_;
 };
@@ -497,8 +501,11 @@ class Assembler : public Malloced {
   void load_rax(void* ptr, RelocInfo::Mode rmode);
   void load_rax(ExternalReference ext);
 
-  // Conditional moves
-  // Implement conditional moves here.
+  // Conditional moves.
+  void cmovq(Condition cc, Register dst, Register src);
+  void cmovq(Condition cc, Register dst, const Operand& src);
+  void cmovl(Condition cc, Register dst, Register src);
+  void cmovl(Condition cc, Register dst, const Operand& src);
 
   // Exchange two registers
   void xchg(Register dst, Register src);
@@ -512,6 +519,10 @@ class Assembler : public Malloced {
     arithmetic_op_32(0x03, dst, src);
   }
 
+  void addl(Register dst, Immediate src) {
+    immediate_arithmetic_op_32(0x0, dst, src);
+  }
+
   void addq(Register dst, const Operand& src) {
     arithmetic_op(0x03, dst, src);
   }
@@ -844,17 +855,32 @@ class Assembler : public Malloced {
 
   void frndint();
 
+  void sahf();
+
   // SSE2 instructions
+  void movsd(const Operand& dst, XMMRegister src);
+  void movsd(Register src, XMMRegister dst);
+  void movsd(XMMRegister dst, Register src);
+  void movsd(XMMRegister src, const Operand& dst);
+
   void cvttss2si(Register dst, const Operand& src);
   void cvttsd2si(Register dst, const Operand& src);
 
-  void cvtsi2sd(XMMRegister dst, const Operand& src);
+  void cvtlsi2sd(XMMRegister dst, const Operand& src);
+  void cvtlsi2sd(XMMRegister dst, Register src);
+  void cvtqsi2sd(XMMRegister dst, const Operand& src);
+  void cvtqsi2sd(XMMRegister dst, Register src);
 
   void addsd(XMMRegister dst, XMMRegister src);
   void subsd(XMMRegister dst, XMMRegister src);
   void mulsd(XMMRegister dst, XMMRegister src);
   void divsd(XMMRegister dst, XMMRegister src);
 
+
+  void emit_sse_operand(XMMRegister dst, XMMRegister src);
+  void emit_sse_operand(XMMRegister reg, const Operand& adr);
+  void emit_sse_operand(XMMRegister dst, Register src);
+
   // Use either movsd or movlpd.
   // void movdbl(XMMRegister dst, const Operand& src);
   // void movdbl(const Operand& dst, XMMRegister src);
@@ -933,6 +959,7 @@ class Assembler : public Malloced {
   // High bit of reg goes to REX.R, high bit of rm_reg goes to REX.B.
   // REX.W is set.
   inline void emit_rex_64(Register reg, Register rm_reg);
+  inline void emit_rex_64(XMMRegister reg, Register rm_reg);
 
   // Emits a REX prefix that encodes a 64-bit operand size and
   // the top bit of the destination, index, and base register codes.
@@ -940,6 +967,7 @@ class Assembler : public Malloced {
   // register is used for REX.B, and the high bit of op's index register
   // is used for REX.X.  REX.W is set.
   inline void emit_rex_64(Register reg, const Operand& op);
+  inline void emit_rex_64(XMMRegister reg, const Operand& op);
 
   // Emits a REX prefix that encodes a 64-bit operand size and
   // the top bit of the register code.
@@ -984,6 +1012,18 @@ class Assembler : public Malloced {
   // is emitted.
   inline void emit_optional_rex_32(Register reg, const Operand& op);
 
+  // As for emit_optional_rex_32(Register, Register), except that
+  // the registers are XMM registers.
+  inline void emit_optional_rex_32(XMMRegister reg, XMMRegister base);
+
+  // As for emit_optional_rex_32(Register, Register), except that
+  // the registers are XMM registers.
+  inline void emit_optional_rex_32(XMMRegister reg, Register base);
+
+  // As for emit_optional_rex_32(Register, const Operand&), except that
+  // the register is an XMM register.
+  inline void emit_optional_rex_32(XMMRegister reg, const Operand& op);
+
   // Optionally do as emit_rex_32(Register) if the register number has
   // the high bit set.
   inline void emit_optional_rex_32(Register rm_reg);

src/x64/macro-assembler-x64.cc

@@ -331,7 +331,15 @@ void MacroAssembler::Jump(Handle<Code> code_object, RelocInfo::Mode rmode) {
   WriteRecordedPositions();
   ASSERT(RelocInfo::IsCodeTarget(rmode));
   movq(kScratchRegister, code_object, rmode);
+#ifdef DEBUG
+  Label target;
+  bind(&target);
+#endif
   jmp(kScratchRegister);
+#ifdef DEBUG
+  ASSERT_EQ(kTargetAddrToReturnAddrDist,
+            SizeOfCodeGeneratedSince(&target) + kPointerSize);
+#endif
 }
 
 
@@ -544,6 +552,36 @@ void MacroAssembler::CopyRegistersFromStackToMemory(Register base,
 #endif  // ENABLE_DEBUGGER_SUPPORT
 
 
+void MacroAssembler::InvokeBuiltin(Builtins::JavaScript id, InvokeFlag flag) {
+  bool resolved;
+  Handle<Code> code = ResolveBuiltin(id, &resolved);
+
+  // Calls are not allowed in some stubs.
+  ASSERT(flag == JUMP_FUNCTION || allow_stub_calls());
+
+  // Rely on the assertion to check that the number of provided
+  // arguments match the expected number of arguments. Fake a
+  // parameter count to avoid emitting code to do the check.
+  ParameterCount expected(0);
+  InvokeCode(Handle<Code>(code), expected, expected,
+             RelocInfo::CODE_TARGET, flag);
+
+  const char* name = Builtins::GetName(id);
+  int argc = Builtins::GetArgumentsCount(id);
+  // The target address for the jump is stored as an immediate at offset
+  // kInvokeCodeAddressOffset.
+  if (!resolved) {
+    uint32_t flags =
+        Bootstrapper::FixupFlagsArgumentsCount::encode(argc) |
+        Bootstrapper::FixupFlagsIsPCRelative::encode(true) |
+        Bootstrapper::FixupFlagsUseCodeObject::encode(false);
+    Unresolved entry =
+        { pc_offset() - kTargetAddrToReturnAddrDist, flags, name };
+    unresolved_.Add(entry);
+  }
+}
+
+
 void MacroAssembler::InvokePrologue(const ParameterCount& expected,
                                     const ParameterCount& actual,
                                     Handle<Code> code_constant,
@@ -610,8 +648,6 @@ void MacroAssembler::InvokePrologue(const ParameterCount& expected,
 }
 
 
-
-
 void MacroAssembler::InvokeCode(Register code,
                                 const ParameterCount& expected,
                                 const ParameterCount& actual,
@@ -636,12 +672,11 @@ void MacroAssembler::InvokeCode(Handle<Code> code,
   Label done;
   Register dummy = rax;
   InvokePrologue(expected, actual, code, dummy, &done, flag);
-  movq(kScratchRegister, code, rmode);
   if (flag == CALL_FUNCTION) {
-    call(kScratchRegister);
+    Call(code, rmode);
   } else {
     ASSERT(flag == JUMP_FUNCTION);
-    jmp(kScratchRegister);
+    Jump(code, rmode);
   }
   bind(&done);
 }

src/x64/stub-cache-x64.cc

@@ -31,12 +31,12 @@
 #include "ic-inl.h"
 #include "codegen-inl.h"
 #include "stub-cache.h"
+#include "macro-assembler-x64.h"
 
 namespace v8 {
 namespace internal {
 
-
-#define __ ACCESS_MASM(masm())
+#define __ ACCESS_MASM((&masm_))
 
 
 Object* CallStubCompiler::CompileCallConstant(Object* a,
@@ -148,7 +148,6 @@ Object* StubCompiler::CompileLazyCompile(Code::Flags flags) {
   return GetCodeWithFlags(flags, "LazyCompileStub");
 }
 
-
 #undef __
 
 

src/x64/virtual-frame-x64.cc

@@ -809,6 +809,19 @@ void VirtualFrame::SyncRange(int begin, int end) {
   }
 }
 
+
+Result VirtualFrame::InvokeBuiltin(Builtins::JavaScript id,
+                                   InvokeFlag flag,
+                                   int arg_count) {
+  PrepareForCall(arg_count, arg_count);
+  ASSERT(cgen()->HasValidEntryRegisters());
+  __ InvokeBuiltin(id, flag);
+  Result result = cgen()->allocator()->Allocate(rax);
+  ASSERT(result.is_valid());
+  return result;
+}
+
+
 //------------------------------------------------------------------------------
 // Virtual frame stub and IC calling functions.