Change clamping 0..255 instruction sequence for pixel array code.

The subb instruction added to the IA-32 assembler is not used as dec_b ended up being used instead.

There is a mesurable difference.
Review URL: http://codereview.chromium.org/246076

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/ia32/assembler-ia32.cc

@@ -1166,6 +1166,19 @@ void Assembler::shr_cl(Register dst) {
 }
 
 
+void Assembler::subb(const Operand& op, int8_t imm8) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  if (op.is_reg(eax)) {
+    EMIT(0x2c);
+  } else {
+    EMIT(0x80);
+    emit_operand(ebp, op);  // ebp == 5
+  }
+  EMIT(imm8);
+}
+
+
 void Assembler::sub(const Operand& dst, const Immediate& x) {
   EnsureSpace ensure_space(this);
   last_pc_ = pc_;

src/ia32/assembler-ia32.h

@@ -590,6 +590,7 @@ class Assembler : public Malloced {
   void shr(Register dst);
   void shr_cl(Register dst);
 
+  void subb(const Operand& dst, int8_t imm8);
   void sub(const Operand& dst, const Immediate& x);
   void sub(Register dst, const Operand& src);
   void sub(const Operand& dst, Register src);

src/ia32/ic-ia32.cc

@@ -298,7 +298,6 @@ void KeyedLoadIC::GenerateGeneric(MacroAssembler* masm) {
   __ shl(eax, kSmiTagSize);
   __ ret(0);
 
-
   // Slow case: Load name and receiver from stack and jump to runtime.
   __ bind(&slow);
   __ IncrementCounter(&Counters::keyed_load_generic_slow, 1);
@@ -427,11 +426,8 @@ void KeyedStoreIC::GenerateGeneric(MacroAssembler* masm) {
     Label done, is_negative;
     __ test(eax, Immediate(0xFFFFFF00));
     __ j(zero, &done);
-    __ j(negative, &is_negative);
-    __ mov(eax, Immediate(255));
-    __ jmp(&done);
-    __ bind(&is_negative);
-    __ xor_(eax, Operand(eax));  // Clear eax.
+    __ setcc(negative, eax);  // 1 if negative, 0 if positive.
+    __ dec_b(eax);  // 0 if negative, 255 if positive.
     __ bind(&done);
   }
   __ mov(ecx, FieldOperand(ecx, PixelArray::kExternalPointerOffset));
@@ -458,7 +454,6 @@ void KeyedStoreIC::GenerateGeneric(MacroAssembler* masm) {
   __ sub(Operand(ebx), Immediate(1 << kSmiTagSize));  // decrement ebx again
   __ jmp(&fast);
 
-
   // Array case: Get the length and the elements array from the JS
   // array. Check that the array is in fast mode; if it is the
   // length is always a smi.

src/x64/assembler-x64.cc

@@ -913,6 +913,27 @@ void Assembler::decl(const Operand& dst) {
 }
 
 
+void Assembler::decb(Register dst) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  if (dst.code() > 3) {
+    // Register is not one of al, bl, cl, dl.  Its encoding needs REX.
+    emit_rex_32(dst);
+  }
+  emit(0xFE);
+  emit_modrm(0x1, dst);
+}
+
+
+void Assembler::decb(const Operand& dst) {
+  EnsureSpace ensure_space(this);
+  last_pc_ = pc_;
+  emit_optional_rex_32(dst);
+  emit(0xFE);
+  emit_operand(1, dst);
+}
+
+
 void Assembler::enter(Immediate size) {
   EnsureSpace ensure_space(this);
   last_pc_ = pc_;

src/x64/assembler-x64.h

@@ -703,6 +703,8 @@ class Assembler : public Malloced {
   void decq(const Operand& dst);
   void decl(Register dst);
   void decl(const Operand& dst);
+  void decb(Register dst);
+  void decb(const Operand& dst);
 
   // Sign-extends rax into rdx:rax.
   void cqo();

src/x64/ic-x64.cc

@@ -491,11 +491,8 @@ void KeyedStoreIC::GenerateGeneric(MacroAssembler* masm) {
     Label done, is_negative;
     __ testl(rax, Immediate(0xFFFFFF00));
     __ j(zero, &done);
-    __ j(negative, &is_negative);
-    __ movl(rax, Immediate(255));
-    __ jmp(&done);
-    __ bind(&is_negative);
-    __ xorl(rax, rax);  // Clear rax.
+    __ setcc(negative, rax);  // 1 if negative, 0 if positive.
+    __ decb(rax);  // 0 if negative, 255 if positive.
     __ bind(&done);
   }
   __ movq(rcx, FieldOperand(rcx, PixelArray::kExternalPointerOffset));