First draft of the sh4 port

src/sh4/code-stubs-sh4.cc

-// Copyright 2012 the V8 project authors. All rights reserved.
+// Copyright 2011-2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
 //       with the distribution.
 //     * Neither the name of Google Inc. nor the names of its
 //       contributors may be used to endorse or promote products derived
 //       from this software without specific prior written permission.
 //
 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "v8.h"
 
-#if defined(V8_TARGET_ARCH_ARM)
+#if defined(V8_TARGET_ARCH_SH4)
 
 #include "bootstrapper.h"
 #include "code-stubs.h"
 #include "regexp-macro-assembler.h"
 
 namespace v8 {
 namespace internal {
 
 
 #define __ ACCESS_MASM(masm)
 
 static void EmitIdenticalObjectComparison(MacroAssembler* masm,
                                           Label* slow,
                                           Condition cond,
                                           bool never_nan_nan);
 static void EmitSmiNonsmiComparison(MacroAssembler* masm,
                                     Register lhs,
                                     Register rhs,
                                     Label* lhs_not_nan,
                                     Label* slow,
                                     bool strict);
 static void EmitTwoNonNanDoubleComparison(MacroAssembler* masm, Condition cond);
 static void EmitStrictTwoHeapObjectCompare(MacroAssembler* masm,
                                            Register lhs,
                                            Register rhs);
+static void EmitCheckForTwoHeapNumbers(MacroAssembler* masm,
+                                       Register lhs,
+                                       Register rhs,
+                                       Label* both_loaded_as_doubles,
+                                       Label* not_heap_numbers,
+                                       Label* slow);
+static void EmitCheckForSymbolsOrObjects(MacroAssembler* masm,
+                                         Register lhs,
+                                         Register rhs,
+                                         Label* possible_strings,
+                                         Label* not_both_strings);
+void EmitNanCheck(MacroAssembler* masm, Label* lhs_not_nan, Condition cond);
 
 
+// Copy from ARM
+#include "map-sh4.h"  // Define register map
 // Check if the operand is a heap number.
 static void EmitCheckForHeapNumber(MacroAssembler* masm, Register operand,
                                    Register scratch1, Register scratch2,
                                    Label* not_a_heap_number) {
   __ ldr(scratch1, FieldMemOperand(operand, HeapObject::kMapOffset));
   __ LoadRoot(scratch2, Heap::kHeapNumberMapRootIndex);
   __ cmp(scratch1, scratch2);
   __ b(ne, not_a_heap_number);
 }
 
 
 void ToNumberStub::Generate(MacroAssembler* masm) {
   // The ToNumber stub takes one argument in eax.
+  // Entry argument: r0
+  // Exit in: r0
+#ifdef DEBUG
+  // Clobber other parameter registers on entry.
+  __ Dead(r1, r2, r3);
+  __ Dead(r4, r5, r6, r7);
+#endif
   Label check_heap_number, call_builtin;
-  __ JumpIfNotSmi(r0, &check_heap_number);
+  __ JumpIfNotSmi(r0, &check_heap_number, Label::kNear);
   __ Ret();
 
   __ bind(&check_heap_number);
   EmitCheckForHeapNumber(masm, r0, r1, ip, &call_builtin);
   __ Ret();
 
   __ bind(&call_builtin);
   __ push(r0);
   __ InvokeBuiltin(Builtins::TO_NUMBER, JUMP_FUNCTION);
 }
@@ skipping 80 matching lines @@
   Label loop;
   __ ldr(r4, FieldMemOperand(r1, FixedArray::kLengthOffset));
   __ bind(&loop);
   // Do not double check first entry.
 
   __ cmp(r4, Operand(Smi::FromInt(SharedFunctionInfo::kEntryLength)));
   __ b(eq, &install_unoptimized);
   __ sub(r4, r4, Operand(
       Smi::FromInt(SharedFunctionInfo::kEntryLength)));  // Skip an entry.
   __ add(r5, r1, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
-  __ add(r5, r5, Operand(r4, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(ip, r4, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r5, r5, ip);
   __ ldr(r5, MemOperand(r5));
   __ cmp(r2, r5);
   __ b(ne, &loop);
   // Hit: fetch the optimized code.
   __ add(r5, r1, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
-  __ add(r5, r5, Operand(r4, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r4, r4, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r5, r5, r4);
   __ add(r5, r5, Operand(kPointerSize));
   __ ldr(r4, MemOperand(r5));
 
   __ bind(&install_optimized);
   __ IncrementCounter(counters->fast_new_closure_install_optimized(),
                       1, r6, r7);
 
   // TODO(fschneider): Idea: store proper code pointers in the map and either
   // unmangle them on marking or do nothing as the whole map is discarded on
   // major GC anyway.
@@ skipping 195 matching lines @@
   // [sp]: constant elements.
   // [sp + kPointerSize]: literal index.
   // [sp + (2 * kPointerSize)]: literals array.
 
   // Load boilerplate object into r3 and check if we need to create a
   // boilerplate.
   Label slow_case;
   __ ldr(r3, MemOperand(sp, 2 * kPointerSize));
   __ ldr(r0, MemOperand(sp, 1 * kPointerSize));
   __ add(r3, r3, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
-  __ ldr(r3, MemOperand(r3, r0, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(ip, r0, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ ldr(r3, MemOperand(r3, ip));
   __ CompareRoot(r3, Heap::kUndefinedValueRootIndex);
   __ b(eq, &slow_case);
 
   FastCloneShallowArrayStub::Mode mode = mode_;
   if (mode == CLONE_ANY_ELEMENTS) {
     Label double_elements, check_fast_elements;
     __ ldr(r0, FieldMemOperand(r3, JSArray::kElementsOffset));
     __ ldr(r0, FieldMemOperand(r0, HeapObject::kMapOffset));
     __ CompareRoot(r0, Heap::kFixedCOWArrayMapRootIndex);
     __ b(ne, &check_fast_elements);
@@ skipping 57 matching lines @@
   // [sp + kPointerSize]: constant properties.
   // [sp + (2 * kPointerSize)]: literal index.
   // [sp + (3 * kPointerSize)]: literals array.
 
   // Load boilerplate object into r3 and check if we need to create a
   // boilerplate.
   Label slow_case;
   __ ldr(r3, MemOperand(sp, 3 * kPointerSize));
   __ ldr(r0, MemOperand(sp, 2 * kPointerSize));
   __ add(r3, r3, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
-  __ ldr(r3, MemOperand(r3, r0, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r1, r0, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ ldr(r3, MemOperand(r3, r1));
   __ CompareRoot(r3, Heap::kUndefinedValueRootIndex);
   __ b(eq, &slow_case);
 
   // Check that the boilerplate contains only fast properties and we can
   // statically determine the instance size.
   int size = JSObject::kHeaderSize + length_ * kPointerSize;
   __ ldr(r0, FieldMemOperand(r3, HeapObject::kMapOffset));
   __ ldrb(r0, FieldMemOperand(r0, Map::kInstanceSizeOffset));
   __ cmp(r0, Operand(size >> kPointerSizeLog2));
   __ b(ne, &slow_case);
@@ skipping 48 matching lines @@
            (result2_.code() << 4) +
            (source_.code() << 8) +
            (zeros_.code() << 12);
   }
 
   void Generate(MacroAssembler* masm);
 };
 
 
 void ConvertToDoubleStub::Generate(MacroAssembler* masm) {
+  ASSERT(!result1_.is(ip) && !result2_.is(ip) && !zeros_.is(ip));
   Register exponent = result1_;
   Register mantissa = result2_;
 
   Label not_special;
   // Convert from Smi to integer.
-  __ mov(source_, Operand(source_, ASR, kSmiTagSize));
+  __ asr(source_, source_, Operand(kSmiTagSize));
   // Move sign bit from source to destination.  This works because the sign bit
   // in the exponent word of the double has the same position and polarity as
   // the 2's complement sign bit in a Smi.
   STATIC_ASSERT(HeapNumber::kSignMask == 0x80000000u);
-  __ and_(exponent, source_, Operand(HeapNumber::kSignMask), SetCC);
+  __ land(exponent, source_, Operand(HeapNumber::kSignMask));
+  __ tst(exponent, exponent);
   // Subtract from 0 if source was negative.
-  __ rsb(source_, source_, Operand(0, RelocInfo::NONE), LeaveCC, ne);
+  __ rsb(source_, source_, Operand(0), ne);
 
   // We have -1, 0 or 1, which we treat specially. Register source_ contains
   // absolute value: it is either equal to 1 (special case of -1 and 1),
   // greater than 1 (not a special case) or less than 1 (special case of 0).
-  __ cmp(source_, Operand(1));
-  __ b(gt, &not_special);
+  __ cmpgt(source_, Operand(1));
+  __ bt_near(&not_special);
 
   // For 1 or -1 we need to or in the 0 exponent (biased to 1023).
   const uint32_t exponent_word_for_1 =
       HeapNumber::kExponentBias << HeapNumber::kExponentShift;
-  __ orr(exponent, exponent, Operand(exponent_word_for_1), LeaveCC, eq);
+  __ cmpeq(source_, Operand(1));
+  __ orr(exponent, exponent, Operand(exponent_word_for_1), eq);
   // 1, 0 and -1 all have 0 for the second word.
   __ mov(mantissa, Operand(0, RelocInfo::NONE));
   __ Ret();
 
   __ bind(&not_special);
   // Count leading zeros.  Uses mantissa for a scratch register on pre-ARM5.
   // Gets the wrong answer for 0, but we already checked for that case above.
   __ CountLeadingZeros(zeros_, source_, mantissa);
   // Compute exponent and or it into the exponent register.
   // We use mantissa as a scratch register here.  Use a fudge factor to
   // divide the constant 31 + HeapNumber::kExponentBias, 0x41d, into two parts
   // that fit in the ARM's constant field.
   int fudge = 0x400;
   __ rsb(mantissa, zeros_, Operand(31 + HeapNumber::kExponentBias - fudge));
   __ add(mantissa, mantissa, Operand(fudge));
-  __ orr(exponent,
-         exponent,
-         Operand(mantissa, LSL, HeapNumber::kExponentShift));
+  __ lsl(ip, mantissa, Operand(HeapNumber::kExponentShift));
+  __ orr(exponent, exponent, ip);
   // Shift up the source chopping the top bit off.
   __ add(zeros_, zeros_, Operand(1));
   // This wouldn't work for 1.0 or -1.0 as the shift would be 32 which means 0.
-  __ mov(source_, Operand(source_, LSL, zeros_));
+  __ lsl(source_, source_, zeros_);
   // Compute lower part of fraction (last 12 bits).
-  __ mov(mantissa, Operand(source_, LSL, HeapNumber::kMantissaBitsInTopWord));
+  __ lsl(mantissa, source_, Operand(HeapNumber::kMantissaBitsInTopWord));
   // And the top (top 20 bits).
-  __ orr(exponent,
-         exponent,
-         Operand(source_, LSR, 32 - HeapNumber::kMantissaBitsInTopWord));
+  __ lsr(ip, source_, Operand(32 - HeapNumber::kMantissaBitsInTopWord));
+  __ orr(exponent, exponent, ip);
   __ Ret();
 }
 
 
 void FloatingPointHelper::LoadSmis(MacroAssembler* masm,
                                    FloatingPointHelper::Destination destination,
                                    Register scratch1,
                                    Register scratch2) {
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    __ mov(scratch1, Operand(r0, ASR, kSmiTagSize));
-    __ vmov(d7.high(), scratch1);
-    __ vcvt_f64_s32(d7, d7.high());
-    __ mov(scratch1, Operand(r1, ASR, kSmiTagSize));
-    __ vmov(d6.high(), scratch1);
-    __ vcvt_f64_s32(d6, d6.high());
+  if (CpuFeatures::IsSupported(FPU)) {
+    __ asr(scratch1, r0, Operand(kSmiTagSize));
+    __ dfloat(dr2, scratch1);
+    __ asr(scratch1, r1, Operand(kSmiTagSize));
+    __ dfloat(dr0, scratch1);
     if (destination == kCoreRegisters) {
-      __ vmov(r2, r3, d7);
-      __ vmov(r0, r1, d6);
+      __ movd(r2, r3, dr2);
+      __ movd(r0, r1, dr0);
     }
   } else {
     ASSERT(destination == kCoreRegisters);
     // Write Smi from r0 to r3 and r2 in double format.
-    __ mov(scratch1, Operand(r0));
+    __ mov(scratch1, r0);
     ConvertToDoubleStub stub1(r3, r2, scratch1, scratch2);
     __ push(lr);
     __ Call(stub1.GetCode());
     // Write Smi from r1 to r1 and r0 in double format.
-    __ mov(scratch1, Operand(r1));
+    __ mov(scratch1, r1);
     ConvertToDoubleStub stub2(r1, r0, scratch1, scratch2);
     __ Call(stub2.GetCode());
     __ pop(lr);
   }
 }
 
 
 void FloatingPointHelper::LoadOperands(
     MacroAssembler* masm,
     FloatingPointHelper::Destination destination,
     Register heap_number_map,
     Register scratch1,
     Register scratch2,
     Label* slow) {
 
-  // Load right operand (r0) to d6 or r2/r3.
+  // Load right operand (r0) to d7 or r2/r3.
+  ASSERT(!heap_number_map.is(r0) && !heap_number_map.is(r1) &&
+         !heap_number_map.is(r2) && !heap_number_map.is(r3));
   LoadNumber(masm, destination,
-             r0, d7, r2, r3, heap_number_map, scratch1, scratch2, slow);
+             r0, dr2, r2, r3, heap_number_map, scratch1, scratch2, slow);
 
-  // Load left operand (r1) to d7 or r0/r1.
+  // Load left operand (r1) to d6 or r0/r1.
   LoadNumber(masm, destination,
-             r1, d6, r0, r1, heap_number_map, scratch1, scratch2, slow);
+             r1, dr0, r0, r1, heap_number_map, scratch1, scratch2, slow);
 }
 
 
 void FloatingPointHelper::LoadNumber(MacroAssembler* masm,
                                      Destination destination,
                                      Register object,
                                      DwVfpRegister dst,
                                      Register dst1,
                                      Register dst2,
                                      Register heap_number_map,
                                      Register scratch1,
                                      Register scratch2,
                                      Label* not_number) {
-  __ AssertRootValue(heap_number_map,
-                     Heap::kHeapNumberMapRootIndex,
-                     "HeapNumberMap register clobbered.");
+  if (FLAG_debug_code) {
+    __ AbortIfNotRootValue(heap_number_map,
+                           Heap::kHeapNumberMapRootIndex,
+                           "HeapNumberMap register clobbered.");
+  }
 
   Label is_smi, done;
 
   // Smi-check
   __ UntagAndJumpIfSmi(scratch1, object, &is_smi);
   // Heap number check
   __ JumpIfNotHeapNumber(object, heap_number_map, scratch1, not_number);
 
   // Handle loading a double from a heap number.
-  if (CpuFeatures::IsSupported(VFP2) &&
+  if (CpuFeatures::IsSupported(FPU) &&
       destination == kVFPRegisters) {
-    CpuFeatures::Scope scope(VFP2);
     // Load the double from tagged HeapNumber to double register.
-    __ sub(scratch1, object, Operand(kHeapObjectTag));
-    __ vldr(dst, scratch1, HeapNumber::kValueOffset);
+    ASSERT(Operand(kHeapObjectTag - HeapNumber::kValueOffset).is_int8());
+    __ sub(scratch1, object, Operand(kHeapObjectTag -
+                                     HeapNumber::kValueOffset));
+    __ dldr(dst, MemOperand(scratch1, 0), scratch1);
   } else {
     ASSERT(destination == kCoreRegisters);
     // Load the double from heap number to dst1 and dst2 in double format.
     __ Ldrd(dst1, dst2, FieldMemOperand(object, HeapNumber::kValueOffset));
   }
-  __ jmp(&done);
+  __ jmp_near(&done);
 
   // Handle loading a double from a smi.
   __ bind(&is_smi);
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    // Convert smi to double using VFP instructions.
-    __ vmov(dst.high(), scratch1);
-    __ vcvt_f64_s32(dst, dst.high());
+  if (CpuFeatures::IsSupported(FPU)) {
+    // Convert smi to double using FPU instructions.
+    __ SmiUntag(scratch1, object);
+    __ dfloat(dst, scratch1);
     if (destination == kCoreRegisters) {
       // Load the converted smi to dst1 and dst2 in double format.
-      __ vmov(dst1, dst2, dst);
+      __ movd(dst1, dst2, dst);
     }
   } else {
     ASSERT(destination == kCoreRegisters);
     // Write smi to dst1 and dst2 double format.
-    __ mov(scratch1, Operand(object));
+    __ mov(scratch1, object);
     ConvertToDoubleStub stub(dst2, dst1, scratch1, scratch2);
     __ push(lr);
     __ Call(stub.GetCode());
     __ pop(lr);
   }
 
   __ bind(&done);
 }
 
 
 void FloatingPointHelper::ConvertNumberToInt32(MacroAssembler* masm,
                                                Register object,
                                                Register dst,
                                                Register heap_number_map,
                                                Register scratch1,
                                                Register scratch2,
                                                Register scratch3,
                                                DwVfpRegister double_scratch,
                                                Label* not_number) {
-  __ AssertRootValue(heap_number_map,
-                     Heap::kHeapNumberMapRootIndex,
-                     "HeapNumberMap register clobbered.");
+  if (FLAG_debug_code) {
+    __ AbortIfNotRootValue(heap_number_map,
+                           Heap::kHeapNumberMapRootIndex,
+                           "HeapNumberMap register clobbered.");
+  }
   Label done;
   Label not_in_int32_range;
 
   __ UntagAndJumpIfSmi(dst, object, &done);
   __ ldr(scratch1, FieldMemOperand(object, HeapNumber::kMapOffset));
   __ cmp(scratch1, heap_number_map);
   __ b(ne, not_number);
   __ ConvertToInt32(object,
                     dst,
                     scratch1,
@@ skipping 21 matching lines @@
                                              Register dst1,
                                              Register dst2,
                                              Register scratch2,
                                              SwVfpRegister single_scratch) {
   ASSERT(!int_scratch.is(scratch2));
   ASSERT(!int_scratch.is(dst1));
   ASSERT(!int_scratch.is(dst2));
 
   Label done;
 
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    __ vmov(single_scratch, int_scratch);
-    __ vcvt_f64_s32(double_dst, single_scratch);
+  if (CpuFeatures::IsSupported(FPU)) {
+    __ dfloat(double_dst, int_scratch);
     if (destination == kCoreRegisters) {
-      __ vmov(dst1, dst2, double_dst);
+      __ movd(dst1, dst2, double_dst);
     }
   } else {
     Label fewer_than_20_useful_bits;
     // Expected output:
     // |         dst2            |         dst1            |
     // | s |   exp   |              mantissa               |
 
     // Check for zero.
-    __ cmp(int_scratch, Operand::Zero());
+    __ cmp(int_scratch, Operand(0));
     __ mov(dst2, int_scratch);
     __ mov(dst1, int_scratch);
     __ b(eq, &done);
 
     // Preload the sign of the value.
-    __ and_(dst2, int_scratch, Operand(HeapNumber::kSignMask), SetCC);
+    __ land(dst2, int_scratch, Operand(HeapNumber::kSignMask));
     // Get the absolute value of the object (as an unsigned integer).
-    __ rsb(int_scratch, int_scratch, Operand::Zero(), SetCC, mi);
+    __ cmpge(dst2, Operand(0));
+    __ rsb(ip, int_scratch, Operand(0));
+    __ mov(int_scratch, ip, f);
 
-    // Get mantissa[51:20].
+    // Get mantisssa[51:20].
 
     // Get the position of the first set bit.
     __ CountLeadingZeros(dst1, int_scratch, scratch2);
     __ rsb(dst1, dst1, Operand(31));
 
     // Set the exponent.
     __ add(scratch2, dst1, Operand(HeapNumber::kExponentBias));
     __ Bfi(dst2, scratch2, scratch2,
         HeapNumber::kExponentShift, HeapNumber::kExponentBits);
 
     // Clear the first non null bit.
     __ mov(scratch2, Operand(1));
-    __ bic(int_scratch, int_scratch, Operand(scratch2, LSL, dst1));
+    __ lsl(scratch2, scratch2, dst1);
+    __ bic(int_scratch, int_scratch, scratch2);
 
-    __ cmp(dst1, Operand(HeapNumber::kMantissaBitsInTopWord));
+    // Present on ARM, but dead code.
+    // __ cmp(dst1, Operand(HeapNumber::kMantissaBitsInTopWord));
     // Get the number of bits to set in the lower part of the mantissa.
-    __ sub(scratch2, dst1, Operand(HeapNumber::kMantissaBitsInTopWord), SetCC);
-    __ b(mi, &fewer_than_20_useful_bits);
+    __ sub(scratch2, dst1, Operand(HeapNumber::kMantissaBitsInTopWord));
+    __ cmpge(scratch2, Operand(0));
+    __ b(f, &fewer_than_20_useful_bits, Label::kNear);
     // Set the higher 20 bits of the mantissa.
-    __ orr(dst2, dst2, Operand(int_scratch, LSR, scratch2));
+    __ lsr(ip, int_scratch, scratch2);
+    __ orr(dst2, dst2, ip);
     __ rsb(scratch2, scratch2, Operand(32));
-    __ mov(dst1, Operand(int_scratch, LSL, scratch2));
+    __ lsl(dst1, int_scratch, scratch2);
     __ b(&done);
 
     __ bind(&fewer_than_20_useful_bits);
     __ rsb(scratch2, dst1, Operand(HeapNumber::kMantissaBitsInTopWord));
-    __ mov(scratch2, Operand(int_scratch, LSL, scratch2));
+    __ lsl(scratch2, int_scratch, scratch2);
     __ orr(dst2, dst2, scratch2);
     // Set dst1 to 0.
-    __ mov(dst1, Operand::Zero());
+    __ mov(dst1, Operand(0));
   }
   __ bind(&done);
 }
 
 
 void FloatingPointHelper::LoadNumberAsInt32Double(MacroAssembler* masm,
                                                   Register object,
                                                   Destination destination,
                                                   DwVfpRegister double_dst,
-                                                  DwVfpRegister double_scratch,
                                                   Register dst1,
                                                   Register dst2,
                                                   Register heap_number_map,
                                                   Register scratch1,
                                                   Register scratch2,
                                                   SwVfpRegister single_scratch,
                                                   Label* not_int32) {
   ASSERT(!scratch1.is(object) && !scratch2.is(object));
   ASSERT(!scratch1.is(scratch2));
   ASSERT(!heap_number_map.is(object) &&
          !heap_number_map.is(scratch1) &&
          !heap_number_map.is(scratch2));
 
   Label done, obj_is_not_smi;
 
   __ JumpIfNotSmi(object, &obj_is_not_smi);
   __ SmiUntag(scratch1, object);
   ConvertIntToDouble(masm, scratch1, destination, double_dst, dst1, dst2,
                      scratch2, single_scratch);
   __ b(&done);
 
   __ bind(&obj_is_not_smi);
-  __ AssertRootValue(heap_number_map,
-                     Heap::kHeapNumberMapRootIndex,
-                     "HeapNumberMap register clobbered.");
+  if (FLAG_debug_code) {
+    __ AbortIfNotRootValue(heap_number_map,
+                           Heap::kHeapNumberMapRootIndex,
+                           "HeapNumberMap register clobbered.");
+  }
   __ JumpIfNotHeapNumber(object, heap_number_map, scratch1, not_int32);
 
   // Load the number.
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
+  if (CpuFeatures::IsSupported(FPU)) {
     // Load the double value.
     __ sub(scratch1, object, Operand(kHeapObjectTag));
-    __ vldr(double_dst, scratch1, HeapNumber::kValueOffset);
+    __ dldr(double_dst, MemOperand(scratch1, HeapNumber::kValueOffset));
 
-    __ EmitVFPTruncate(kRoundToZero,
+    __ EmitFPUTruncate(kRoundToZero,
                        scratch1,
                        double_dst,
                        scratch2,
-                       double_scratch,
                        kCheckForInexactConversion);
 
     // Jump to not_int32 if the operation did not succeed.
     __ b(ne, not_int32);
 
     if (destination == kCoreRegisters) {
-      __ vmov(dst1, dst2, double_dst);
+      __ movd(dst1, dst2, double_dst);
     }
 
   } else {
     ASSERT(!scratch1.is(object) && !scratch2.is(object));
     // Load the double value in the destination registers..
     __ Ldrd(dst1, dst2, FieldMemOperand(object, HeapNumber::kValueOffset));
 
     // Check for 0 and -0.
     __ bic(scratch1, dst1, Operand(HeapNumber::kSignMask));
-    __ orr(scratch1, scratch1, Operand(dst2));
-    __ cmp(scratch1, Operand::Zero());
+    __ orr(scratch1, scratch1, dst2);
+    __ cmp(scratch1, Operand(0));
     __ b(eq, &done);
 
     // Check that the value can be exactly represented by a 32-bit integer.
     // Jump to not_int32 if that's not the case.
     DoubleIs32BitInteger(masm, dst1, dst2, scratch1, scratch2, not_int32);
 
     // dst1 and dst2 were trashed. Reload the double value.
     __ Ldrd(dst1, dst2, FieldMemOperand(object, HeapNumber::kValueOffset));
   }
 
   __ bind(&done);
 }
 
 
 void FloatingPointHelper::LoadNumberAsInt32(MacroAssembler* masm,
                                             Register object,
                                             Register dst,
                                             Register heap_number_map,
                                             Register scratch1,
                                             Register scratch2,
                                             Register scratch3,
-                                            DwVfpRegister double_scratch0,
-                                            DwVfpRegister double_scratch1,
+                                            DwVfpRegister double_scratch,
                                             Label* not_int32) {
   ASSERT(!dst.is(object));
   ASSERT(!scratch1.is(object) && !scratch2.is(object) && !scratch3.is(object));
   ASSERT(!scratch1.is(scratch2) &&
          !scratch1.is(scratch3) &&
          !scratch2.is(scratch3));
 
   Label done;
 
   __ UntagAndJumpIfSmi(dst, object, &done);
 
-  __ AssertRootValue(heap_number_map,
-                     Heap::kHeapNumberMapRootIndex,
-                     "HeapNumberMap register clobbered.");
+  if (FLAG_debug_code) {
+    __ AbortIfNotRootValue(heap_number_map,
+                           Heap::kHeapNumberMapRootIndex,
+                           "HeapNumberMap register clobbered.");
+  }
   __ JumpIfNotHeapNumber(object, heap_number_map, scratch1, not_int32);
 
   // Object is a heap number.
   // Convert the floating point value to a 32-bit integer.
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-
+  if (CpuFeatures::IsSupported(FPU)) {
     // Load the double value.
     __ sub(scratch1, object, Operand(kHeapObjectTag));
-    __ vldr(double_scratch0, scratch1, HeapNumber::kValueOffset);
+    __ dldr(double_scratch, MemOperand(scratch1, HeapNumber::kValueOffset));
 
-    __ EmitVFPTruncate(kRoundToZero,
-                       dst,
-                       double_scratch0,
+    __ EmitFPUTruncate(kRoundToZero,
+                       scratch2,
+                       double_scratch,
                        scratch1,
-                       double_scratch1,
                        kCheckForInexactConversion);
 
     // Jump to not_int32 if the operation did not succeed.
     __ b(ne, not_int32);
+    // Get the result in the destination register.
+    __ mov(dst, scratch2);
+
   } else {
     // Load the double value in the destination registers.
     __ ldr(scratch1, FieldMemOperand(object, HeapNumber::kExponentOffset));
     __ ldr(scratch2, FieldMemOperand(object, HeapNumber::kMantissaOffset));
 
     // Check for 0 and -0.
     __ bic(dst, scratch1, Operand(HeapNumber::kSignMask));
-    __ orr(dst, scratch2, Operand(dst));
-    __ cmp(dst, Operand::Zero());
+    __ orr(dst, scratch2, dst);
+    __ cmp(dst, Operand(0));
     __ b(eq, &done);
 
     DoubleIs32BitInteger(masm, scratch1, scratch2, dst, scratch3, not_int32);
 
     // Registers state after DoubleIs32BitInteger.
     // dst: mantissa[51:20].
     // scratch2: 1
 
     // Shift back the higher bits of the mantissa.
-    __ mov(dst, Operand(dst, LSR, scratch3));
+    __ lsr(dst, dst, scratch3);
     // Set the implicit first bit.
     __ rsb(scratch3, scratch3, Operand(32));
-    __ orr(dst, dst, Operand(scratch2, LSL, scratch3));
+    __ lsl(scratch2, scratch2, scratch3);
+    __ orr(dst, dst, scratch2);
     // Set the sign.
     __ ldr(scratch1, FieldMemOperand(object, HeapNumber::kExponentOffset));
     __ tst(scratch1, Operand(HeapNumber::kSignMask));
-    __ rsb(dst, dst, Operand::Zero(), LeaveCC, mi);
+    __ rsb(ip, dst, Operand(0));
+    __ mov(dst, ip, ne);        // FIXME(stm): strange case !!
   }
 
   __ bind(&done);
 }
 
 
 void FloatingPointHelper::DoubleIs32BitInteger(MacroAssembler* masm,
                                                Register src1,
                                                Register src2,
                                                Register dst,
                                                Register scratch,
                                                Label* not_int32) {
   // Get exponent alone in scratch.
   __ Ubfx(scratch,
           src1,
           HeapNumber::kExponentShift,
           HeapNumber::kExponentBits);
 
   // Substract the bias from the exponent.
-  __ sub(scratch, scratch, Operand(HeapNumber::kExponentBias), SetCC);
+  __ sub(scratch, scratch, Operand(HeapNumber::kExponentBias));
 
   // src1: higher (exponent) part of the double value.
   // src2: lower (mantissa) part of the double value.
   // scratch: unbiased exponent.
 
   // Fast cases. Check for obvious non 32-bit integer values.
   // Negative exponent cannot yield 32-bit integers.
-  __ b(mi, not_int32);
+  __ cmpge(scratch, Operand(0));
+  __ b(f, not_int32);
   // Exponent greater than 31 cannot yield 32-bit integers.
   // Also, a positive value with an exponent equal to 31 is outside of the
   // signed 32-bit integer range.
   // Another way to put it is that if (exponent - signbit) > 30 then the
   // number cannot be represented as an int32.
   Register tmp = dst;
-  __ sub(tmp, scratch, Operand(src1, LSR, 31));
-  __ cmp(tmp, Operand(30));
-  __ b(gt, not_int32);
+  __ lsr(tmp, src1, Operand(31));
+  __ sub(tmp, scratch, tmp);
+  __ cmpgt(tmp, Operand(30));
+  __ b(t, not_int32);
   // - Bits [21:0] in the mantissa are not null.
   __ tst(src2, Operand(0x3fffff));
   __ b(ne, not_int32);
 
   // Otherwise the exponent needs to be big enough to shift left all the
   // non zero bits left. So we need the (30 - exponent) last bits of the
   // 31 higher bits of the mantissa to be null.
   // Because bits [21:0] are null, we can check instead that the
-  // (32 - exponent) last bits of the 32 higher bits of the mantissa are null.
+  // (32 - exponent) last bits of the 32 higher bits of the mantisssa are null.
 
   // Get the 32 higher bits of the mantissa in dst.
   __ Ubfx(dst,
           src2,
           HeapNumber::kMantissaBitsInTopWord,
           32 - HeapNumber::kMantissaBitsInTopWord);
+  __ lsl(ip, src1, Operand(HeapNumber::kNonMantissaBitsInTopWord));
   __ orr(dst,
          dst,
-         Operand(src1, LSL, HeapNumber::kNonMantissaBitsInTopWord));
+         ip);
 
   // Create the mask and test the lower bits (of the higher bits).
   __ rsb(scratch, scratch, Operand(32));
   __ mov(src2, Operand(1));
-  __ mov(src1, Operand(src2, LSL, scratch));
+  __ lsl(src1, src2, scratch);
   __ sub(src1, src1, Operand(1));
   __ tst(dst, src1);
   __ b(ne, not_int32);
 }
 
 
 void FloatingPointHelper::CallCCodeForDoubleOperation(
     MacroAssembler* masm,
     Token::Value op,
     Register heap_number_result,
     Register scratch) {
   // Using core registers:
   // r0: Left value (least significant part of mantissa).
   // r1: Left value (sign, exponent, top of mantissa).
   // r2: Right value (least significant part of mantissa).
   // r3: Right value (sign, exponent, top of mantissa).
 
   // Assert that heap_number_result is callee-saved.
   // We currently always use r5 to pass it.
+  // Note: as r5 is not callee-saved on SH4, we push/pop it below
   ASSERT(heap_number_result.is(r5));
 
+  // Calling C function (using double registers): move r0..r3 to fr4..fr7
+  __ movd(dr4, r0, r1);
+  __ movd(dr6, r2, r3);
+
   // Push the current return address before the C call. Return will be
   // through pop(pc) below.
   __ push(lr);
-  __ PrepareCallCFunction(0, 2, scratch);
-  if (masm->use_eabi_hardfloat()) {
-    CpuFeatures::Scope scope(VFP2);
-    __ vmov(d0, r0, r1);
-    __ vmov(d1, r2, r3);
-  }
+  __ push(heap_number_result);  // sh4 specific
+  /* Use r0 as scratch: PrepareCallCFunction() disallow use of r4-r7 on sh4. */
+  __ PrepareCallCFunction(0, 2, r0);
   {
     AllowExternalCallThatCantCauseGC scope(masm);
     __ CallCFunction(
         ExternalReference::double_fp_operation(op, masm->isolate()), 0, 2);
   }
+  __ movd(r0, r1, dr0);
   // Store answer in the overwritable heap number. Double returned in
   // registers r0 and r1 or in d0.
-  if (masm->use_eabi_hardfloat()) {
-    CpuFeatures::Scope scope(VFP2);
-    __ vstr(d0,
-            FieldMemOperand(heap_number_result, HeapNumber::kValueOffset));
-  } else {
-    __ Strd(r0, r1, FieldMemOperand(heap_number_result,
-                                    HeapNumber::kValueOffset));
-  }
+  __ pop(heap_number_result);  // sh4 specific
+  __ Strd(r0, r1, FieldMemOperand(heap_number_result,
+                                  HeapNumber::kValueOffset));
   // Place heap_number_result in r0 and return to the pushed return address.
-  __ mov(r0, Operand(heap_number_result));
-  __ pop(pc);
+  __ mov(r0, heap_number_result);
+  __ pop(lr);
+  __ rts();
 }
 
 
 bool WriteInt32ToHeapNumberStub::IsPregenerated() {
   // These variants are compiled ahead of time.  See next method.
   if (the_int_.is(r1) && the_heap_number_.is(r0) && scratch_.is(r2)) {
     return true;
   }
   if (the_int_.is(r2) && the_heap_number_.is(r0) && scratch_.is(r3)) {
     return true;
   }
   // Other register combinations are generated as and when they are needed,
   // so it is unsafe to call them from stubs (we can't generate a stub while
   // we are generating a stub).
   return false;
 }
 
 
 void WriteInt32ToHeapNumberStub::GenerateFixedRegStubsAheadOfTime() {
   WriteInt32ToHeapNumberStub stub1(r1, r0, r2);
   WriteInt32ToHeapNumberStub stub2(r2, r0, r3);
   stub1.GetCode()->set_is_pregenerated(true);
   stub2.GetCode()->set_is_pregenerated(true);
 }
 
 
 // See comment for class.
 void WriteInt32ToHeapNumberStub::Generate(MacroAssembler* masm) {
+  ASSERT(!scratch_.is(ip) && !the_int_.is(ip));
   Label max_negative_int;
   // the_int_ has the answer which is a signed int32 but not a Smi.
   // We test for the special value that has a different exponent.  This test
   // has the neat side effect of setting the flags according to the sign.
   STATIC_ASSERT(HeapNumber::kSignMask == 0x80000000u);
   __ cmp(the_int_, Operand(0x80000000u));
-  __ b(eq, &max_negative_int);
+  __ b(eq, &max_negative_int, Label::kNear);
   // Set up the correct exponent in scratch_.  All non-Smi int32s have the same.
   // A non-Smi integer is 1.xxx * 2^30 so the exponent is 30 (biased).
   uint32_t non_smi_exponent =
       (HeapNumber::kExponentBias + 30) << HeapNumber::kExponentShift;
   __ mov(scratch_, Operand(non_smi_exponent));
+  __ cmpge(the_int_, Operand(0));
+  Label skip;
+  __ bt_near(&skip);
   // Set the sign bit in scratch_ if the value was negative.
-  __ orr(scratch_, scratch_, Operand(HeapNumber::kSignMask), LeaveCC, cs);
+  __ lor(scratch_, scratch_, Operand(HeapNumber::kSignMask));
   // Subtract from 0 if the value was negative.
-  __ rsb(the_int_, the_int_, Operand(0, RelocInfo::NONE), LeaveCC, cs);
+  __ rsb(the_int_, the_int_, Operand(0));
+  __ bind(&skip);
   // We should be masking the implict first digit of the mantissa away here,
   // but it just ends up combining harmlessly with the last digit of the
   // exponent that happens to be 1.  The sign bit is 0 so we shift 10 to get
   // the most significant 1 to hit the last bit of the 12 bit sign and exponent.
   ASSERT(((1 << HeapNumber::kExponentShift) & non_smi_exponent) != 0);
   const int shift_distance = HeapNumber::kNonMantissaBitsInTopWord - 2;
-  __ orr(scratch_, scratch_, Operand(the_int_, LSR, shift_distance));
+  __ lsr(ip, the_int_, Operand(shift_distance));
+  __ lor(scratch_, scratch_, ip);
   __ str(scratch_, FieldMemOperand(the_heap_number_,
                                    HeapNumber::kExponentOffset));
-  __ mov(scratch_, Operand(the_int_, LSL, 32 - shift_distance));
+  __ lsl(scratch_, the_int_, Operand(32 - shift_distance));
   __ str(scratch_, FieldMemOperand(the_heap_number_,
                                    HeapNumber::kMantissaOffset));
   __ Ret();
 
   __ bind(&max_negative_int);
   // The max negative int32 is stored as a positive number in the mantissa of
   // a double because it uses a sign bit instead of using two's complement.
   // The actual mantissa bits stored are all 0 because the implicit most
   // significant 1 bit is not stored.
   non_smi_exponent += 1 << HeapNumber::kExponentShift;
@@ skipping 18 matching lines @@
   __ b(ne, &not_identical);
 
   // The two objects are identical.  If we know that one of them isn't NaN then
   // we now know they test equal.
   if (cond != eq || !never_nan_nan) {
     // Test for NaN. Sadly, we can't just compare to FACTORY->nan_value(),
     // so we do the second best thing - test it ourselves.
     // They are both equal and they are not both Smis so both of them are not
     // Smis.  If it's not a heap number, then return equal.
     if (cond == lt || cond == gt) {
-      __ CompareObjectType(r0, r4, r4, FIRST_SPEC_OBJECT_TYPE);
-      __ b(ge, slow);
+      __ CompareObjectType(r0, r4, r4, FIRST_SPEC_OBJECT_TYPE, ge);
+      __ bt(slow);
     } else {
-      __ CompareObjectType(r0, r4, r4, HEAP_NUMBER_TYPE);
-      __ b(eq, &heap_number);
+      __ CompareObjectType(r0, r4, r4, HEAP_NUMBER_TYPE, eq);
+      __ b(eq, &heap_number, Label::kNear);
       // Comparing JS objects with <=, >= is complicated.
       if (cond != eq) {
-        __ cmp(r4, Operand(FIRST_SPEC_OBJECT_TYPE));
-        __ b(ge, slow);
+        __ cmpge(r4, Operand(FIRST_SPEC_OBJECT_TYPE));
+        __ bt(slow);
         // Normally here we fall through to return_equal, but undefined is
         // special: (undefined == undefined) == true, but
         // (undefined <= undefined) == false!  See ECMAScript 11.8.5.
         if (cond == le || cond == ge) {
           __ cmp(r4, Operand(ODDBALL_TYPE));
-          __ b(ne, &return_equal);
+          __ b(ne, &return_equal, Label::kNear);
           __ LoadRoot(r2, Heap::kUndefinedValueRootIndex);
           __ cmp(r0, r2);
-          __ b(ne, &return_equal);
+          __ b(ne, &return_equal, Label::kNear);
           if (cond == le) {
             // undefined <= undefined should fail.
             __ mov(r0, Operand(GREATER));
           } else  {
             // undefined >= undefined should fail.
             __ mov(r0, Operand(LESS));
           }
           __ Ret();
         }
       }
@@ skipping 23 matching lines @@
       // and not all mantissa bits (0..51) clear.
       // Read top bits of double representation (second word of value).
       __ ldr(r2, FieldMemOperand(r0, HeapNumber::kExponentOffset));
       // Test that exponent bits are all set.
       __ Sbfx(r3, r2, HeapNumber::kExponentShift, HeapNumber::kExponentBits);
       // NaNs have all-one exponents so they sign extend to -1.
       __ cmp(r3, Operand(-1));
       __ b(ne, &return_equal);
 
       // Shift out flag and all exponent bits, retaining only mantissa.
-      __ mov(r2, Operand(r2, LSL, HeapNumber::kNonMantissaBitsInTopWord));
+      __ lsl(r2, r2, Operand(HeapNumber::kNonMantissaBitsInTopWord));
       // Or with all low-bits of mantissa.
       __ ldr(r3, FieldMemOperand(r0, HeapNumber::kMantissaOffset));
-      __ orr(r0, r3, Operand(r2), SetCC);
+      __ orr(r0, r3, r2);
+      __ tst(r0, r0);
       // For equal we already have the right value in r0:  Return zero (equal)
       // if all bits in mantissa are zero (it's an Infinity) and non-zero if
       // not (it's a NaN).  For <= and >= we need to load r0 with the failing
       // value if it's a NaN.
       if (cond != eq) {
         // All-zero means Infinity means equal.
         __ Ret(eq);
         if (cond == le) {
           __ mov(r0, Operand(GREATER));  // NaN <= NaN should fail.
         } else {
@@ skipping 13 matching lines @@
 static void EmitSmiNonsmiComparison(MacroAssembler* masm,
                                     Register lhs,
                                     Register rhs,
                                     Label* lhs_not_nan,
                                     Label* slow,
                                     bool strict) {
   ASSERT((lhs.is(r0) && rhs.is(r1)) ||
          (lhs.is(r1) && rhs.is(r0)));
 
   Label rhs_is_smi;
-  __ JumpIfSmi(rhs, &rhs_is_smi);
+  __ JumpIfSmi(rhs, &rhs_is_smi, Label::kNear);
 
   // Lhs is a Smi.  Check whether the rhs is a heap number.
-  __ CompareObjectType(rhs, r4, r4, HEAP_NUMBER_TYPE);
+  __ CompareObjectType(rhs, r4, r4, HEAP_NUMBER_TYPE, eq);
   if (strict) {
     // If rhs is not a number and lhs is a Smi then strict equality cannot
     // succeed.  Return non-equal
     // If rhs is r0 then there is already a non zero value in it.
     if (!rhs.is(r0)) {
-      __ mov(r0, Operand(NOT_EQUAL), LeaveCC, ne);
+      __ mov(r0, Operand(NOT_EQUAL), ne);
     }
     __ Ret(ne);
   } else {
     // Smi compared non-strictly with a non-Smi non-heap-number.  Call
     // the runtime.
     __ b(ne, slow);
   }
 
   // Lhs is a smi, rhs is a number.
-  if (CpuFeatures::IsSupported(VFP2)) {
-    // Convert lhs to a double in d7.
-    CpuFeatures::Scope scope(VFP2);
-    __ SmiToDoubleVFPRegister(lhs, d7, r7, s15);
-    // Load the double from rhs, tagged HeapNumber r0, to d6.
-    __ sub(r7, rhs, Operand(kHeapObjectTag));
-    __ vldr(d6, r7, HeapNumber::kValueOffset);
+  if (CpuFeatures::IsSupported(FPU)) {
+    // Convert lhs to a double in dr2.
+    __ SmiToDoubleFPURegister(lhs, dr2, r7);
+    // Load the double from rhs, tagged HeapNumber r0, to dr0.
+    __ sub(r7, rhs, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dldr(dr0, MemOperand(r7, 0), r7);
   } else {
     __ push(lr);
     // Convert lhs to a double in r2, r3.
-    __ mov(r7, Operand(lhs));
+    __ mov(r7, lhs);
     ConvertToDoubleStub stub1(r3, r2, r7, r6);
     __ Call(stub1.GetCode());
     // Load rhs to a double in r0, r1.
     __ Ldrd(r0, r1, FieldMemOperand(rhs, HeapNumber::kValueOffset));
     __ pop(lr);
   }
 
   // We now have both loaded as doubles but we can skip the lhs nan check
   // since it's a smi.
   __ jmp(lhs_not_nan);
 
   __ bind(&rhs_is_smi);
   // Rhs is a smi.  Check whether the non-smi lhs is a heap number.
-  __ CompareObjectType(lhs, r4, r4, HEAP_NUMBER_TYPE);
+  __ CompareObjectType(lhs, r4, r4, HEAP_NUMBER_TYPE, eq);
   if (strict) {
     // If lhs is not a number and rhs is a smi then strict equality cannot
     // succeed.  Return non-equal.
     // If lhs is r0 then there is already a non zero value in it.
     if (!lhs.is(r0)) {
-      __ mov(r0, Operand(NOT_EQUAL), LeaveCC, ne);
+      __ mov(r0, Operand(NOT_EQUAL), ne);
     }
     __ Ret(ne);
   } else {
     // Smi compared non-strictly with a non-smi non-heap-number.  Call
     // the runtime.
     __ b(ne, slow);
   }
 
   // Rhs is a smi, lhs is a heap number.
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    // Load the double from lhs, tagged HeapNumber r1, to d7.
-    __ sub(r7, lhs, Operand(kHeapObjectTag));
-    __ vldr(d7, r7, HeapNumber::kValueOffset);
-    // Convert rhs to a double in d6              .
-    __ SmiToDoubleVFPRegister(rhs, d6, r7, s13);
+  if (CpuFeatures::IsSupported(FPU)) {
+    // Load the double from lhs, tagged HeapNumber r1, to dr2.
+    __ sub(r7, lhs, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dldr(dr2, MemOperand(r7, 0), r7);
+    // Convert rhs to a double in dr0.
+    __ SmiToDoubleFPURegister(rhs, dr0, r7);
   } else {
     __ push(lr);
     // Load lhs to a double in r2, r3.
     __ Ldrd(r2, r3, FieldMemOperand(lhs, HeapNumber::kValueOffset));
     // Convert rhs to a double in r0, r1.
-    __ mov(r7, Operand(rhs));
+    __ mov(r7, rhs);
     ConvertToDoubleStub stub2(r1, r0, r7, r6);
     __ Call(stub2.GetCode());
     __ pop(lr);
   }
   // Fall through to both_loaded_as_doubles.
 }
 
 
 void EmitNanCheck(MacroAssembler* masm, Label* lhs_not_nan, Condition cond) {
   bool exp_first = (HeapNumber::kExponentOffset == HeapNumber::kValueOffset);
   Register rhs_exponent = exp_first ? r0 : r1;
   Register lhs_exponent = exp_first ? r2 : r3;
   Register rhs_mantissa = exp_first ? r1 : r0;
   Register lhs_mantissa = exp_first ? r3 : r2;
   Label one_is_nan, neither_is_nan;
 
   __ Sbfx(r4,
           lhs_exponent,
           HeapNumber::kExponentShift,
           HeapNumber::kExponentBits);
   // NaNs have all-one exponents so they sign extend to -1.
   __ cmp(r4, Operand(-1));
   __ b(ne, lhs_not_nan);
-  __ mov(r4,
-         Operand(lhs_exponent, LSL, HeapNumber::kNonMantissaBitsInTopWord),
-         SetCC);
-  __ b(ne, &one_is_nan);
+  __ lsl(r4, lhs_exponent, Operand(HeapNumber::kNonMantissaBitsInTopWord));
+  __ cmpeq(r4, Operand(0));
+  __ b(ne, &one_is_nan, Label::kNear);
   __ cmp(lhs_mantissa, Operand(0, RelocInfo::NONE));
-  __ b(ne, &one_is_nan);
+  __ b(ne, &one_is_nan, Label::kNear);
 
   __ bind(lhs_not_nan);
   __ Sbfx(r4,
           rhs_exponent,
           HeapNumber::kExponentShift,
           HeapNumber::kExponentBits);
   // NaNs have all-one exponents so they sign extend to -1.
   __ cmp(r4, Operand(-1));
-  __ b(ne, &neither_is_nan);
-  __ mov(r4,
-         Operand(rhs_exponent, LSL, HeapNumber::kNonMantissaBitsInTopWord),
-         SetCC);
-  __ b(ne, &one_is_nan);
-  __ cmp(rhs_mantissa, Operand(0, RelocInfo::NONE));
-  __ b(eq, &neither_is_nan);
+  __ b(ne, &neither_is_nan, Label::kNear);
+  __ lsl(r4, rhs_exponent, Operand(HeapNumber::kNonMantissaBitsInTopWord));
+  __ cmpeq(r4, Operand(0, RelocInfo::NONE));
+  __ b(ne, &one_is_nan, Label::kNear);
+  __ cmp(rhs_mantissa, Operand(0));
+  __ b(eq, &neither_is_nan, Label::kNear);
 
   __ bind(&one_is_nan);
   // NaN comparisons always fail.
   // Load whatever we need in r0 to make the comparison fail.
   if (cond == lt || cond == le) {
     __ mov(r0, Operand(GREATER));
   } else {
     __ mov(r0, Operand(LESS));
   }
   __ Ret();
 
   __ bind(&neither_is_nan);
 }
 
 
 // See comment at call site.
 static void EmitTwoNonNanDoubleComparison(MacroAssembler* masm,
                                           Condition cond) {
   bool exp_first = (HeapNumber::kExponentOffset == HeapNumber::kValueOffset);
   Register rhs_exponent = exp_first ? r0 : r1;
   Register lhs_exponent = exp_first ? r2 : r3;
   Register rhs_mantissa = exp_first ? r1 : r0;
   Register lhs_mantissa = exp_first ? r3 : r2;
 
   // r0, r1, r2, r3 have the two doubles.  Neither is a NaN.
   if (cond == eq) {
     // Doubles are not equal unless they have the same bit pattern.
     // Exception: 0 and -0.
-    __ cmp(rhs_mantissa, Operand(lhs_mantissa));
-    __ orr(r0, rhs_mantissa, Operand(lhs_mantissa), LeaveCC, ne);
+    __ cmp(rhs_mantissa, lhs_mantissa);
+    __ orr(r0, rhs_mantissa, lhs_mantissa, ne);
     // Return non-zero if the numbers are unequal.
     __ Ret(ne);
 
-    __ sub(r0, rhs_exponent, Operand(lhs_exponent), SetCC);
+    __ sub(r0, rhs_exponent, lhs_exponent);
+    __ tst(r0, r0);
     // If exponents are equal then return 0.
     __ Ret(eq);
 
     // Exponents are unequal.  The only way we can return that the numbers
     // are equal is if one is -0 and the other is 0.  We already dealt
     // with the case where both are -0 or both are 0.
     // We start by seeing if the mantissas (that are equal) or the bottom
     // 31 bits of the rhs exponent are non-zero.  If so we return not
     // equal.
-    __ orr(r4, lhs_mantissa, Operand(lhs_exponent, LSL, kSmiTagSize), SetCC);
-    __ mov(r0, Operand(r4), LeaveCC, ne);
+    __ lsl(r4, lhs_exponent, Operand(kSmiTagSize));
+    __ orr(r4, lhs_mantissa, r4);
+    __ tst(r4, r4);
+    __ mov(r0, r4, ne);
     __ Ret(ne);
     // Now they are equal if and only if the lhs exponent is zero in its
     // low 31 bits.
-    __ mov(r0, Operand(rhs_exponent, LSL, kSmiTagSize));
+    __ lsl(r0, rhs_exponent, Operand(kSmiTagSize));
     __ Ret();
   } else {
+    __ Push(r4, r5, r6, r7);
+    // Calling C function: move r0..r3 to fr4..fr7
+    __ movd(dr4, r0, r1);
+    __ movd(dr6, r2, r3);
+
     // Call a native function to do a comparison between two non-NaNs.
     // Call C routine that may not cause GC or other trouble.
     __ push(lr);
-    __ PrepareCallCFunction(0, 2, r5);
-    if (masm->use_eabi_hardfloat()) {
-      CpuFeatures::Scope scope(VFP2);
-      __ vmov(d0, r0, r1);
-      __ vmov(d1, r2, r3);
-    }
-
-    AllowExternalCallThatCantCauseGC scope(masm);
+    __ PrepareCallCFunction(0, 2, r0);
     __ CallCFunction(ExternalReference::compare_doubles(masm->isolate()),
                      0, 2);
-    __ pop(pc);  // Return.
+    __ pop(lr);
+    __ Pop(r4, r5, r6, r7);
+    __ Ret();
   }
 }
 
 
 // See comment at call site.
 static void EmitStrictTwoHeapObjectCompare(MacroAssembler* masm,
                                            Register lhs,
                                            Register rhs) {
     ASSERT((lhs.is(r0) && rhs.is(r1)) ||
            (lhs.is(r1) && rhs.is(r0)));
 
     // If either operand is a JS object or an oddball value, then they are
     // not equal since their pointers are different.
     // There is no test for undetectability in strict equality.
     STATIC_ASSERT(LAST_TYPE == LAST_SPEC_OBJECT_TYPE);
     Label first_non_object;
     // Get the type of the first operand into r2 and compare it with
     // FIRST_SPEC_OBJECT_TYPE.
-    __ CompareObjectType(rhs, r2, r2, FIRST_SPEC_OBJECT_TYPE);
-    __ b(lt, &first_non_object);
+    __ CompareObjectType(rhs, r2, r2, FIRST_SPEC_OBJECT_TYPE, ge);
+    __ bf_near(&first_non_object);
 
     // Return non-zero (r0 is not zero)
     Label return_not_equal;
     __ bind(&return_not_equal);
     __ Ret();
 
     __ bind(&first_non_object);
     // Check for oddballs: true, false, null, undefined.
     __ cmp(r2, Operand(ODDBALL_TYPE));
     __ b(eq, &return_not_equal);
 
-    __ CompareObjectType(lhs, r3, r3, FIRST_SPEC_OBJECT_TYPE);
-    __ b(ge, &return_not_equal);
+    __ CompareObjectType(lhs, r3, r3, FIRST_SPEC_OBJECT_TYPE, ge);
+    __ bt(&return_not_equal);
 
     // Check for oddballs: true, false, null, undefined.
     __ cmp(r3, Operand(ODDBALL_TYPE));
     __ b(eq, &return_not_equal);
 
     // Now that we have the types we might as well check for symbol-symbol.
     // Ensure that no non-strings have the symbol bit set.
     STATIC_ASSERT(LAST_TYPE < kNotStringTag + kIsSymbolMask);
     STATIC_ASSERT(kSymbolTag != 0);
-    __ and_(r2, r2, Operand(r3));
+    __ land(r2, r2, r3);
     __ tst(r2, Operand(kIsSymbolMask));
     __ b(ne, &return_not_equal);
 }
 
 
 // See comment at call site.
 static void EmitCheckForTwoHeapNumbers(MacroAssembler* masm,
                                        Register lhs,
                                        Register rhs,
                                        Label* both_loaded_as_doubles,
                                        Label* not_heap_numbers,
                                        Label* slow) {
   ASSERT((lhs.is(r0) && rhs.is(r1)) ||
          (lhs.is(r1) && rhs.is(r0)));
 
-  __ CompareObjectType(rhs, r3, r2, HEAP_NUMBER_TYPE);
+  __ CompareObjectType(rhs, r3, r2, HEAP_NUMBER_TYPE, eq);
   __ b(ne, not_heap_numbers);
   __ ldr(r2, FieldMemOperand(lhs, HeapObject::kMapOffset));
   __ cmp(r2, r3);
   __ b(ne, slow);  // First was a heap number, second wasn't.  Go slow case.
 
   // Both are heap numbers.  Load them up then jump to the code we have
   // for that.
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    __ sub(r7, rhs, Operand(kHeapObjectTag));
-    __ vldr(d6, r7, HeapNumber::kValueOffset);
-    __ sub(r7, lhs, Operand(kHeapObjectTag));
-    __ vldr(d7, r7, HeapNumber::kValueOffset);
+  if (CpuFeatures::IsSupported(FPU)) {
+    __ sub(r7, rhs, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dldr(dr0, MemOperand(r7, 0), r7);
+    __ sub(r7, lhs, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dldr(dr2, MemOperand(r7, 0), r7);
   } else {
     __ Ldrd(r2, r3, FieldMemOperand(lhs, HeapNumber::kValueOffset));
     __ Ldrd(r0, r1, FieldMemOperand(rhs, HeapNumber::kValueOffset));
   }
   __ jmp(both_loaded_as_doubles);
 }
 
 
 // Fast negative check for symbol-to-symbol equality.
 static void EmitCheckForSymbolsOrObjects(MacroAssembler* masm,
                                          Register lhs,
                                          Register rhs,
                                          Label* possible_strings,
                                          Label* not_both_strings) {
   ASSERT((lhs.is(r0) && rhs.is(r1)) ||
          (lhs.is(r1) && rhs.is(r0)));
 
   // r2 is object type of rhs.
   // Ensure that no non-strings have the symbol bit set.
   Label object_test;
   STATIC_ASSERT(kSymbolTag != 0);
   __ tst(r2, Operand(kIsNotStringMask));
-  __ b(ne, &object_test);
+  __ b(ne, &object_test, Label::kNear);
   __ tst(r2, Operand(kIsSymbolMask));
   __ b(eq, possible_strings);
-  __ CompareObjectType(lhs, r3, r3, FIRST_NONSTRING_TYPE);
-  __ b(ge, not_both_strings);
+  __ CompareObjectType(lhs, r3, r3, FIRST_NONSTRING_TYPE, ge);
+  __ bt(not_both_strings);
   __ tst(r3, Operand(kIsSymbolMask));
   __ b(eq, possible_strings);
 
   // Both are symbols.  We already checked they weren't the same pointer
   // so they are not equal.
   __ mov(r0, Operand(NOT_EQUAL));
   __ Ret();
 
   __ bind(&object_test);
-  __ cmp(r2, Operand(FIRST_SPEC_OBJECT_TYPE));
-  __ b(lt, not_both_strings);
-  __ CompareObjectType(lhs, r2, r3, FIRST_SPEC_OBJECT_TYPE);
-  __ b(lt, not_both_strings);
+  __ cmpge(r2, Operand(FIRST_SPEC_OBJECT_TYPE));
+  __ bf(not_both_strings);
+  __ CompareObjectType(lhs, r2, r3, FIRST_SPEC_OBJECT_TYPE, ge);
+  __ bf(not_both_strings);
   // If both objects are undetectable, they are equal. Otherwise, they
   // are not equal, since they are different objects and an object is not
   // equal to undefined.
   __ ldr(r3, FieldMemOperand(rhs, HeapObject::kMapOffset));
   __ ldrb(r2, FieldMemOperand(r2, Map::kBitFieldOffset));
   __ ldrb(r3, FieldMemOperand(r3, Map::kBitFieldOffset));
-  __ and_(r0, r2, Operand(r3));
-  __ and_(r0, r0, Operand(1 << Map::kIsUndetectable));
+  __ land(r0, r2, r3);
+  __ land(r0, r0, Operand(1 << Map::kIsUndetectable));
   __ eor(r0, r0, Operand(1 << Map::kIsUndetectable));
   __ Ret();
 }
 
 
 void NumberToStringStub::GenerateLookupNumberStringCache(MacroAssembler* masm,
                                                          Register object,
                                                          Register result,
                                                          Register scratch1,
                                                          Register scratch2,
                                                          Register scratch3,
                                                          bool object_is_smi,
                                                          Label* not_found) {
   // Use of registers. Register result is used as a temporary.
   Register number_string_cache = result;
   Register mask = scratch3;
 
   // Load the number string cache.
   __ LoadRoot(number_string_cache, Heap::kNumberStringCacheRootIndex);
 
   // Make the hash mask from the length of the number string cache. It
   // contains two elements (number and string) for each cache entry.
   __ ldr(mask, FieldMemOperand(number_string_cache, FixedArray::kLengthOffset));
   // Divide length by two (length is a smi).
-  __ mov(mask, Operand(mask, ASR, kSmiTagSize + 1));
+  __ asr(mask, mask, Operand(kSmiTagSize + 1));
   __ sub(mask, mask, Operand(1));  // Make mask.
 
   // Calculate the entry in the number string cache. The hash value in the
   // number string cache for smis is just the smi value, and the hash for
   // doubles is the xor of the upper and lower words. See
   // Heap::GetNumberStringCache.
   Isolate* isolate = masm->isolate();
   Label is_smi;
   Label load_result_from_cache;
   if (!object_is_smi) {
-    __ JumpIfSmi(object, &is_smi);
-    if (CpuFeatures::IsSupported(VFP2)) {
-      CpuFeatures::Scope scope(VFP2);
+    __ JumpIfSmi(object, &is_smi, Label::kNear);
+
+    if (CpuFeatures::IsSupported(FPU)) {
       __ CheckMap(object,
                   scratch1,
                   Heap::kHeapNumberMapRootIndex,
                   not_found,
                   DONT_DO_SMI_CHECK);
 
       STATIC_ASSERT(8 == kDoubleSize);
       __ add(scratch1,
              object,
              Operand(HeapNumber::kValueOffset - kHeapObjectTag));
-      __ ldm(ia, scratch1, scratch1.bit() | scratch2.bit());
-      __ eor(scratch1, scratch1, Operand(scratch2));
-      __ and_(scratch1, scratch1, Operand(mask));
+      __ ldr(scratch2, MemOperand(scratch1, 4));
+      __ ldr(scratch1, MemOperand(scratch1, 0));
+
+      __ eor(scratch1, scratch1, scratch2);
+      __ land(scratch1, scratch1, mask);
 
       // Calculate address of entry in string cache: each entry consists
       // of two pointer sized fields.
-      __ add(scratch1,
-             number_string_cache,
-             Operand(scratch1, LSL, kPointerSizeLog2 + 1));
+      __ lsl(scratch1, scratch1, Operand(kPointerSizeLog2 + 1));
+      __ add(scratch1, number_string_cache, scratch1);
 
       Register probe = mask;
       __ ldr(probe,
              FieldMemOperand(scratch1, FixedArray::kHeaderSize));
       __ JumpIfSmi(probe, not_found);
-      __ sub(scratch2, object, Operand(kHeapObjectTag));
-      __ vldr(d0, scratch2, HeapNumber::kValueOffset);
+      __ sub(scratch2, object, Operand(kHeapObjectTag -
+                                       HeapNumber::kValueOffset));
+      __ dldr(dr0, MemOperand(scratch2, 0), scratch2);
       __ sub(probe, probe, Operand(kHeapObjectTag));
-      __ vldr(d1, probe, HeapNumber::kValueOffset);
-      __ VFPCompareAndSetFlags(d0, d1);
+      __ dldr(dr2, MemOperand(probe, HeapNumber::kValueOffset));
+      __ dcmpeq(dr0, dr2);
       __ b(ne, not_found);  // The cache did not contain this value.
       __ b(&load_result_from_cache);
     } else {
       __ b(not_found);
     }
   }
 
   __ bind(&is_smi);
   Register scratch = scratch1;
-  __ and_(scratch, mask, Operand(object, ASR, 1));
+  __ asr(scratch, object, Operand(1));
+  __ land(scratch, mask, scratch);
   // Calculate address of entry in string cache: each entry consists
   // of two pointer sized fields.
-  __ add(scratch,
-         number_string_cache,
-         Operand(scratch, LSL, kPointerSizeLog2 + 1));
+  __ lsl(scratch, scratch, Operand(kPointerSizeLog2 + 1));
+  __ add(scratch, number_string_cache, scratch);
 
   // Check if the entry is the smi we are looking for.
   Register probe = mask;
   __ ldr(probe, FieldMemOperand(scratch, FixedArray::kHeaderSize));
   __ cmp(object, probe);
   __ b(ne, not_found);
 
   // Get the result from the cache.
   __ bind(&load_result_from_cache);
   __ ldr(result,
          FieldMemOperand(scratch, FixedArray::kHeaderSize + kPointerSize));
   __ IncrementCounter(isolate->counters()->number_to_string_native(),
                       1,
                       scratch1,
                       scratch2);
 }
 
 
 void NumberToStringStub::Generate(MacroAssembler* masm) {
+  // Entry argument: on stack
+  // Exit in: r0
+
   Label runtime;
 
   __ ldr(r1, MemOperand(sp, 0));
 
   // Generate code to lookup number in the number string cache.
   GenerateLookupNumberStringCache(masm, r1, r0, r2, r3, r4, false, &runtime);
   __ add(sp, sp, Operand(1 * kPointerSize));
   __ Ret();
 
   __ bind(&runtime);
   // Handle number to string in the runtime system if not found in the cache.
   __ TailCallRuntime(Runtime::kNumberToStringSkipCache, 1, 1);
 }
 
 
 // On entry lhs_ and rhs_ are the values to be compared.
 // On exit r0 is 0, positive or negative to indicate the result of
 // the comparison.
 void CompareStub::Generate(MacroAssembler* masm) {
   ASSERT((lhs_.is(r0) && rhs_.is(r1)) ||
          (lhs_.is(r1) && rhs_.is(r0)));
 
   Label slow;  // Call builtin.
   Label not_smis, both_loaded_as_doubles, lhs_not_nan;
 
   if (include_smi_compare_) {
     Label not_two_smis, smi_done;
     __ orr(r2, r1, r0);
-    __ JumpIfNotSmi(r2, &not_two_smis);
-    __ mov(r1, Operand(r1, ASR, 1));
-    __ sub(r0, r1, Operand(r0, ASR, 1));
+    __ JumpIfNotSmi(r2, &not_two_smis, Label::kNear);
+    __ asr(r1, r1, Operand(1));
+    __ asr(r0, r0, Operand(1));
+    __ sub(r0, r1, r0);
     __ Ret();
     __ bind(&not_two_smis);
   } else if (FLAG_debug_code) {
     __ orr(r2, r1, r0);
     __ tst(r2, Operand(kSmiTagMask));
     __ Assert(ne, "CompareStub: unexpected smi operands.");
   }
 
   // NOTICE! This code is only reached after a smi-fast-case check, so
   // it is certain that at least one operand isn't a smi.
 
   // Handle the case where the objects are identical.  Either returns the answer
   // or goes to slow.  Only falls through if the objects were not identical.
   EmitIdenticalObjectComparison(masm, &slow, cc_, never_nan_nan_);
 
   // If either is a Smi (we know that not both are), then they can only
   // be strictly equal if the other is a HeapNumber.
   STATIC_ASSERT(kSmiTag == 0);
   ASSERT_EQ(0, Smi::FromInt(0));
-  __ and_(r2, lhs_, Operand(rhs_));
+  __ land(r2, lhs_, rhs_);
   __ JumpIfNotSmi(r2, &not_smis);
   // One operand is a smi.  EmitSmiNonsmiComparison generates code that can:
   // 1) Return the answer.
   // 2) Go to slow.
   // 3) Fall through to both_loaded_as_doubles.
   // 4) Jump to lhs_not_nan.
   // In cases 3 and 4 we have found out we were dealing with a number-number
-  // comparison.  If VFP3 is supported the double values of the numbers have
-  // been loaded into d7 and d6.  Otherwise, the double values have been loaded
-  // into r0, r1, r2, and r3.
+  // comparison.  If FPU is supported the double values of the numbers have
+  // been loaded into dr2 and dr0.  Otherwise, the double values have been
+  // loaded into r0, r1, r2, and r3.
   EmitSmiNonsmiComparison(masm, lhs_, rhs_, &lhs_not_nan, &slow, strict_);
 
   __ bind(&both_loaded_as_doubles);
-  // The arguments have been converted to doubles and stored in d6 and d7, if
-  // VFP3 is supported, or in r0, r1, r2, and r3.
+  // The arguments have been converted to doubles and stored in dr0 and dr2, if
+  // FPU is supported, or in r0, r1, r2, and r3.
   Isolate* isolate = masm->isolate();
-  if (CpuFeatures::IsSupported(VFP2)) {
+  if (CpuFeatures::IsSupported(FPU)) {
     __ bind(&lhs_not_nan);
-    CpuFeatures::Scope scope(VFP2);
-    Label no_nan;
-    // ARMv7 VFP3 instructions to implement double precision comparison.
-    __ VFPCompareAndSetFlags(d7, d6);
+
+    // Test for NaN
     Label nan;
-    __ b(vs, &nan);
-    __ mov(r0, Operand(EQUAL), LeaveCC, eq);
-    __ mov(r0, Operand(LESS), LeaveCC, lt);
-    __ mov(r0, Operand(GREATER), LeaveCC, gt);
-    __ Ret();
+    __ dcmpeq(dr0, dr0);
+    __ bf_near(&nan);
+    __ dcmpeq(dr2, dr2);
+    __ bf_near(&nan);
+
+    // Test for eq, lt and gt
+    Label equal, greater;
+    __ dcmpeq(dr2, dr0);
+    __ bt_near(&equal);
+    __ dcmpgt(dr2, dr0);
+    __ bt_near(&greater);
+
+    __ mov(r0, Operand(LESS));
+    __ rts();
+
+    __ bind(&equal);
+    __ mov(r0, Operand(EQUAL));
+    __ rts();
+
+    __ bind(&greater);
+    __ mov(r0, Operand(GREATER));
+    __ rts();
 
     __ bind(&nan);
-    // If one of the sides was a NaN then the v flag is set.  Load r0 with
-    // whatever it takes to make the comparison fail, since comparisons with NaN
-    // always fail.
+    // One of the sides was a NaN .Load r0 with whatever it takes to make the
+    // comparison fail, since comparisons with NaN always fail.
     if (cc_ == lt || cc_ == le) {
       __ mov(r0, Operand(GREATER));
     } else {
       __ mov(r0, Operand(LESS));
     }
     __ Ret();
   } else {
     // Checks for NaN in the doubles we have loaded.  Can return the answer or
     // fall through if neither is a NaN.  Also binds lhs_not_nan.
     EmitNanCheck(masm, &lhs_not_nan, cc_);
@@ skipping 114 matching lines @@
     __ JumpIfSmi(tos_, &patch);
   }
 
   if (types_.NeedsMap()) {
     __ ldr(map, FieldMemOperand(tos_, HeapObject::kMapOffset));
 
     if (types_.CanBeUndetectable()) {
       __ ldrb(ip, FieldMemOperand(map, Map::kBitFieldOffset));
       __ tst(ip, Operand(1 << Map::kIsUndetectable));
       // Undetectable -> false.
-      __ mov(tos_, Operand(0, RelocInfo::NONE), LeaveCC, ne);
-      __ Ret(ne);
+      Label skip;
+      __ bt_near(&skip);
+      __ mov(tos_, Operand(0, RelocInfo::NONE));
+      __ rts();
+      __ bind(&skip);
     }
   }
 
   if (types_.Contains(SPEC_OBJECT)) {
     // Spec object -> true.
-    __ CompareInstanceType(map, ip, FIRST_SPEC_OBJECT_TYPE);
+    __ CompareInstanceType(map, ip, FIRST_SPEC_OBJECT_TYPE, ge);
     // tos_ contains the correct non-zero return value already.
-    __ Ret(ge);
+    __ Ret(eq);
   }
 
   if (types_.Contains(STRING)) {
     // String value -> false iff empty.
-  __ CompareInstanceType(map, ip, FIRST_NONSTRING_TYPE);
-  __ ldr(tos_, FieldMemOperand(tos_, String::kLengthOffset), lt);
-  __ Ret(lt);  // the string length is OK as the return value
+  __ CompareInstanceType(map, ip, FIRST_NONSTRING_TYPE, ge);
+  Label skip;
+  __ bt_near(&skip);
+  __ ldr(tos_, FieldMemOperand(tos_, String::kLengthOffset));
+  __ rts();  // the string length is OK as the return value
+  __ bind(&skip);
   }
 
   if (types_.Contains(HEAP_NUMBER)) {
     // Heap number -> false iff +0, -0, or NaN.
     Label not_heap_number;
     __ CompareRoot(map, Heap::kHeapNumberMapRootIndex);
-    __ b(ne, &not_heap_number);
+    __ bf(&not_heap_number);
 
-    if (CpuFeatures::IsSupported(VFP2)) {
-      CpuFeatures::Scope scope(VFP2);
-
-      __ vldr(d1, FieldMemOperand(tos_, HeapNumber::kValueOffset));
-      __ VFPCompareAndSetFlags(d1, 0.0);
-      // "tos_" is a register, and contains a non zero value by default.
-      // Hence we only need to overwrite "tos_" with zero to return false for
-      // FP_ZERO or FP_NAN cases. Otherwise, by default it returns true.
-      __ mov(tos_, Operand(0, RelocInfo::NONE), LeaveCC, eq);  // for FP_ZERO
-      __ mov(tos_, Operand(0, RelocInfo::NONE), LeaveCC, vs);  // for FP_NAN
-    } else {
-      Label done, not_nan, not_zero;
-      __ ldr(temp, FieldMemOperand(tos_, HeapNumber::kExponentOffset));
-      // -0 maps to false:
-      __ bic(
-          temp, temp, Operand(HeapNumber::kSignMask, RelocInfo::NONE), SetCC);
-      __ b(ne, &not_zero);
-      // If exponent word is zero then the answer depends on the mantissa word.
-      __ ldr(tos_, FieldMemOperand(tos_, HeapNumber::kMantissaOffset));
-      __ jmp(&done);
-
-      // Check for NaN.
-      __ bind(&not_zero);
-      // We already zeroed the sign bit, now shift out the mantissa so we only
-      // have the exponent left.
-      __ mov(temp, Operand(temp, LSR, HeapNumber::kMantissaBitsInTopWord));
-      unsigned int shifted_exponent_mask =
-          HeapNumber::kExponentMask >> HeapNumber::kMantissaBitsInTopWord;
-      __ cmp(temp, Operand(shifted_exponent_mask, RelocInfo::NONE));
-      __ b(ne, &not_nan);  // If exponent is not 0x7ff then it can't be a NaN.
-
-      // Reload exponent word.
-      __ ldr(temp, FieldMemOperand(tos_, HeapNumber::kExponentOffset));
-      __ tst(temp, Operand(HeapNumber::kMantissaMask, RelocInfo::NONE));
-      // If mantissa is not zero then we have a NaN, so return 0.
-      __ mov(tos_, Operand(0, RelocInfo::NONE), LeaveCC, ne);
-      __ b(ne, &done);
-
-      // Load mantissa word.
-      __ ldr(temp, FieldMemOperand(tos_, HeapNumber::kMantissaOffset));
-      __ cmp(temp, Operand(0, RelocInfo::NONE));
-      // If mantissa is not zero then we have a NaN, so return 0.
-      __ mov(tos_, Operand(0, RelocInfo::NONE), LeaveCC, ne);
-      __ b(ne, &done);
-
-      __ bind(&not_nan);
-      __ mov(tos_, Operand(1, RelocInfo::NONE));
-      __ bind(&done);
-    }
+  if (CpuFeatures::IsSupported(FPU)) {
+    __ dldr(dr0, FieldMemOperand(tos_, HeapNumber::kValueOffset));
+    // "tos_" is a register, and contains a non zero value by default.
+    // Hence we only need to overwrite "tos_" with zero to return false for
+    // FP_ZERO or FP_NAN cases. Otherwise, by default it returns true.
+    __ dfloat(dr2, Operand(0));
+    __ dcmpeq(dr0, dr2);
+    __ mov(tos_, Operand(0, RelocInfo::NONE), eq);  // for FP_ZERO
+    __ dcmpeq(dr0, dr0);
+    // for FP_NAN (dr0 != dr0 iff isnan(dr0))
+    __ mov(tos_, Operand(0, RelocInfo::NONE), ne);
+  } else {
+    UNIMPLEMENTED();
+  }
     __ Ret();
     __ bind(&not_heap_number);
   }
 
   __ bind(&patch);
   GenerateTypeTransition(masm);
 }
 
 
 void ToBooleanStub::CheckOddball(MacroAssembler* masm,
                                  Type type,
                                  Heap::RootListIndex value,
                                  bool result) {
   if (types_.Contains(type)) {
     // If we see an expected oddball, return its ToBoolean value tos_.
     __ LoadRoot(ip, value);
     __ cmp(tos_, ip);
     // The value of a root is never NULL, so we can avoid loading a non-null
     // value into tos_ when we want to return 'true'.
     if (!result) {
-      __ mov(tos_, Operand(0, RelocInfo::NONE), LeaveCC, eq);
+      __ mov(tos_, Operand(0, RelocInfo::NONE), eq);
     }
     __ Ret(eq);
   }
 }
 
 
 void ToBooleanStub::GenerateTypeTransition(MacroAssembler* masm) {
   if (!tos_.is(r3)) {
-    __ mov(r3, Operand(tos_));
+    __ mov(r3, tos_);
   }
   __ mov(r2, Operand(Smi::FromInt(tos_.code())));
   __ mov(r1, Operand(Smi::FromInt(types_.ToByte())));
   __ Push(r3, r2, r1);
   // Patch the caller to an appropriate specialized stub and return the
   // operation result to the caller of the stub.
   __ TailCallExternalReference(
       ExternalReference(IC_Utility(IC::kToBoolean_Patch), masm->isolate()),
       3,
       1);
 }
 
 
 void StoreBufferOverflowStub::Generate(MacroAssembler* masm) {
   // We don't allow a GC during a store buffer overflow so there is no need to
   // store the registers in any particular way, but we do have to store and
   // restore them.
-  __ stm(db_w, sp, kCallerSaved | lr.bit());
+  __ pushm(kJSCallerSaved);
+  __ push(pr);
   if (save_doubles_ == kSaveFPRegs) {
-    CpuFeatures::Scope scope(VFP2);
-    __ sub(sp, sp, Operand(kDoubleSize * DwVfpRegister::kNumRegisters));
-    for (int i = 0; i < DwVfpRegister::kNumRegisters; i++) {
-      DwVfpRegister reg = DwVfpRegister::from_code(i);
-      __ vstr(reg, MemOperand(sp, i * kDoubleSize));
-    }
+      UNIMPLEMENTED();
   }
   const int argument_count = 1;
   const int fp_argument_count = 0;
   const Register scratch = r1;
 
   AllowExternalCallThatCantCauseGC scope(masm);
   __ PrepareCallCFunction(argument_count, fp_argument_count, scratch);
   __ mov(r0, Operand(ExternalReference::isolate_address()));
   __ CallCFunction(
       ExternalReference::store_buffer_overflow_function(masm->isolate()),
       argument_count);
   if (save_doubles_ == kSaveFPRegs) {
-    CpuFeatures::Scope scope(VFP2);
-    for (int i = 0; i < DwVfpRegister::kNumRegisters; i++) {
-      DwVfpRegister reg = DwVfpRegister::from_code(i);
-      __ vldr(reg, MemOperand(sp, i * kDoubleSize));
-    }
-    __ add(sp, sp, Operand(kDoubleSize * DwVfpRegister::kNumRegisters));
+    UNIMPLEMENTED();
   }
-  __ ldm(ia_w, sp, kCallerSaved | pc.bit());  // Also pop pc to get Ret(0).
+  __ pop(pr);
+  __ popm(kJSCallerSaved);
+  __ rts();
 }
 
 
 void UnaryOpStub::PrintName(StringStream* stream) {
   const char* op_name = Token::Name(op_);
   const char* overwrite_name = NULL;  // Make g++ happy.
   switch (mode_) {
     case UNARY_NO_OVERWRITE: overwrite_name = "Alloc"; break;
     case UNARY_OVERWRITE: overwrite_name = "Overwrite"; break;
   }
@@ skipping 17 matching lines @@
       GenerateHeapNumberStub(masm);
       break;
     case UnaryOpIC::GENERIC:
       GenerateGenericStub(masm);
       break;
   }
 }
 
 
 void UnaryOpStub::GenerateTypeTransition(MacroAssembler* masm) {
-  __ mov(r3, Operand(r0));  // the operand
+  __ mov(r3, r0);  // the operand
   __ mov(r2, Operand(Smi::FromInt(op_)));
   __ mov(r1, Operand(Smi::FromInt(mode_)));
   __ mov(r0, Operand(Smi::FromInt(operand_type_)));
   __ Push(r3, r2, r1, r0);
 
   __ TailCallExternalReference(
       ExternalReference(IC_Utility(IC::kUnaryOp_Patch), masm->isolate()), 4, 1);
 }
 
 
@@ skipping 28 matching lines @@
   GenerateTypeTransition(masm);
 }
 
 
 void UnaryOpStub::GenerateSmiCodeSub(MacroAssembler* masm,
                                      Label* non_smi,
                                      Label* slow) {
   __ JumpIfNotSmi(r0, non_smi);
 
   // The result of negating zero or the smallest negative smi is not a smi.
-  __ bic(ip, r0, Operand(0x80000000), SetCC);
+  __ bic(ip, r0, Operand(0x80000000));
+  __ tst(ip, ip);
   __ b(eq, slow);
 
   // Return '0 - value'.
   __ rsb(r0, r0, Operand(0, RelocInfo::NONE));
   __ Ret();
 }
 
 
 void UnaryOpStub::GenerateSmiCodeBitNot(MacroAssembler* masm,
                                         Label* non_smi) {
   __ JumpIfNotSmi(r0, non_smi);
 
   // Flip bits and revert inverted smi-tag.
-  __ mvn(r0, Operand(r0));
+  __ mvn(r0, r0);
   __ bic(r0, r0, Operand(kSmiTagMask));
   __ Ret();
 }
 
 
 // TODO(svenpanne): Use virtual functions instead of switch.
 void UnaryOpStub::GenerateHeapNumberStub(MacroAssembler* masm) {
   switch (op_) {
     case Token::SUB:
       GenerateHeapNumberStubSub(masm);
@@ skipping 32 matching lines @@
                                             Label* slow) {
   EmitCheckForHeapNumber(masm, r0, r1, r6, slow);
   // r0 is a heap number.  Get a new heap number in r1.
   if (mode_ == UNARY_OVERWRITE) {
     __ ldr(r2, FieldMemOperand(r0, HeapNumber::kExponentOffset));
     __ eor(r2, r2, Operand(HeapNumber::kSignMask));  // Flip sign.
     __ str(r2, FieldMemOperand(r0, HeapNumber::kExponentOffset));
   } else {
     Label slow_allocate_heapnumber, heapnumber_allocated;
     __ AllocateHeapNumber(r1, r2, r3, r6, &slow_allocate_heapnumber);
-    __ jmp(&heapnumber_allocated);
+    __ jmp_near(&heapnumber_allocated);
 
     __ bind(&slow_allocate_heapnumber);
     {
       FrameScope scope(masm, StackFrame::INTERNAL);
       __ push(r0);
       __ CallRuntime(Runtime::kNumberAlloc, 0);
-      __ mov(r1, Operand(r0));
+      __ mov(r1, r0);
       __ pop(r0);
     }
 
     __ bind(&heapnumber_allocated);
     __ ldr(r3, FieldMemOperand(r0, HeapNumber::kMantissaOffset));
     __ ldr(r2, FieldMemOperand(r0, HeapNumber::kExponentOffset));
     __ str(r3, FieldMemOperand(r1, HeapNumber::kMantissaOffset));
     __ eor(r2, r2, Operand(HeapNumber::kSignMask));  // Flip sign.
     __ str(r2, FieldMemOperand(r1, HeapNumber::kExponentOffset));
-    __ mov(r0, Operand(r1));
+    __ mov(r0, r1);
   }
   __ Ret();
 }
 
 
 void UnaryOpStub::GenerateHeapNumberCodeBitNot(
     MacroAssembler* masm, Label* slow) {
   Label impossible;
 
   EmitCheckForHeapNumber(masm, r0, r1, r6, slow);
   // Convert the heap number is r0 to an untagged integer in r1.
-  __ ConvertToInt32(r0, r1, r2, r3, d0, slow);
+  __ ConvertToInt32(r0, r1, r2, r3, dr0, slow);
 
   // Do the bitwise operation and check if the result fits in a smi.
   Label try_float;
-  __ mvn(r1, Operand(r1));
-  __ add(r2, r1, Operand(0x40000000), SetCC);
-  __ b(mi, &try_float);
+  __ mvn(r1, r1);
+  __ add(r2, r1, Operand(0x40000000));
+  __ cmpge(r2, Operand(0));
+  __ bf(&try_float);
 
   // Tag the result as a smi and we're done.
-  __ mov(r0, Operand(r1, LSL, kSmiTagSize));
+  __ lsl(r0, r1, Operand(kSmiTagSize));
   __ Ret();
 
   // Try to store the result in a heap number.
   __ bind(&try_float);
   if (mode_ == UNARY_NO_OVERWRITE) {
     Label slow_allocate_heapnumber, heapnumber_allocated;
     // Allocate a new heap number without zapping r0, which we need if it fails.
     __ AllocateHeapNumber(r2, r3, r4, r6, &slow_allocate_heapnumber);
     __ jmp(&heapnumber_allocated);
 
     __ bind(&slow_allocate_heapnumber);
     {
       FrameScope scope(masm, StackFrame::INTERNAL);
       __ push(r0);  // Push the heap number, not the untagged int32.
       __ CallRuntime(Runtime::kNumberAlloc, 0);
       __ mov(r2, r0);  // Move the new heap number into r2.
       // Get the heap number into r0, now that the new heap number is in r2.
       __ pop(r0);
     }
 
     // Convert the heap number in r0 to an untagged integer in r1.
     // This can't go slow-case because it's the same number we already
     // converted once again.
-    __ ConvertToInt32(r0, r1, r3, r4, d0, &impossible);
-    __ mvn(r1, Operand(r1));
+    __ ConvertToInt32(r0, r1, r3, r4, dr0, &impossible);
+    __ mvn(r1, r1);
 
     __ bind(&heapnumber_allocated);
     __ mov(r0, r2);  // Move newly allocated heap number to r0.
   }
 
-  if (CpuFeatures::IsSupported(VFP2)) {
+  if (CpuFeatures::IsSupported(FPU)) {
     // Convert the int32 in r1 to the heap number in r0. r2 is corrupted.
-    CpuFeatures::Scope scope(VFP2);
-    __ vmov(s0, r1);
-    __ vcvt_f64_s32(d0, s0);
-    __ sub(r2, r0, Operand(kHeapObjectTag));
-    __ vstr(d0, r2, HeapNumber::kValueOffset);
-    __ Ret();
+    __ dfloat(dr0, r1);
+    __ sub(r2, r0, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dstr(dr0, MemOperand(r2, 0), r2);
+    __ rts();
   } else {
     // WriteInt32ToHeapNumberStub does not trigger GC, so we do not
     // have to set up a frame.
     WriteInt32ToHeapNumberStub stub(r1, r0, r2);
     __ Jump(stub.GetCode(), RelocInfo::CODE_TARGET);
   }
 
   __ bind(&impossible);
   if (FLAG_debug_code) {
     __ stop("Incorrect assumption in bit-not stub");
@@ skipping 65 matching lines @@
   __ TailCallExternalReference(
       ExternalReference(IC_Utility(IC::kBinaryOp_Patch),
                         masm->isolate()),
       5,
       1);
 }
 
 
 void BinaryOpStub::GenerateTypeTransitionWithSavedArgs(
     MacroAssembler* masm) {
-  UNIMPLEMENTED();
+  __ UNIMPLEMENTED_BREAK();
 }
 
 
 void BinaryOpStub::Generate(MacroAssembler* masm) {
   // Explicitly allow generation of nested stubs. It is safe here because
   // generation code does not use any raw pointers.
   AllowStubCallsScope allow_stub_calls(masm, true);
 
   switch (operands_type_) {
     case BinaryOpIC::UNINITIALIZED:
@@ skipping 44 matching lines @@
 
 void BinaryOpStub::GenerateSmiSmiOperation(MacroAssembler* masm) {
   Register left = r1;
   Register right = r0;
   Register scratch1 = r7;
   Register scratch2 = r9;
 
   ASSERT(right.is(r0));
   STATIC_ASSERT(kSmiTag == 0);
 
-  Label not_smi_result;
+  Label not_smi_result, skip_if_true, skip_if_false;
   switch (op_) {
     case Token::ADD:
-      __ add(right, left, Operand(right), SetCC);  // Add optimistically.
-      __ Ret(vc);
-      __ sub(right, right, Operand(left));  // Revert optimistic add.
+      __ addv(right, left, right);        // Add optimistically.
+      __ Ret(f);                        // Return if no overflow
+      __ sub(right, right, left);          // Revert optimistic add.
       break;
     case Token::SUB:
-      __ sub(right, left, Operand(right), SetCC);  // Subtract optimistically.
-      __ Ret(vc);
-      __ sub(right, left, Operand(right));  // Revert optimistic subtract.
+      __ subv(right, left, right);        // Subtract optimistically.
+      __ Ret(f);                        // Return if no overflow
+      __ sub(right, left, right);        // Revert optimistic subtract.
       break;
     case Token::MUL:
+      // TODO(stm): implement optimized multiply with overflow check for SH4
       // Remove tag from one of the operands. This way the multiplication result
       // will be a smi if it fits the smi range.
       __ SmiUntag(ip, right);
       // Do multiplication
       // scratch1 = lower 32 bits of ip * left.
       // scratch2 = higher 32 bits of ip * left.
-      __ smull(scratch1, scratch2, left, ip);
+      __ dmuls(scratch1, scratch2, left, ip);
       // Check for overflowing the smi range - no overflow if higher 33 bits of
       // the result are identical.
-      __ mov(ip, Operand(scratch1, ASR, 31));
-      __ cmp(ip, Operand(scratch2));
+      __ asr(ip, scratch1, Operand(31));
+      __ cmp(ip, scratch2);
       __ b(ne, &not_smi_result);
       // Go slow on zero result to handle -0.
-      __ cmp(scratch1, Operand(0));
-      __ mov(right, Operand(scratch1), LeaveCC, ne);
-      __ Ret(ne);
+      __ tst(scratch1, scratch1);
+      __ bt_near(&skip_if_true);
+      __ mov(right, scratch1);
+      __ rts();
+      __ bind(&skip_if_true);
       // We need -0 if we were multiplying a negative number with 0 to get 0.
       // We know one of them was zero.
-      __ add(scratch2, right, Operand(left), SetCC);
-      __ mov(right, Operand(Smi::FromInt(0)), LeaveCC, pl);
-      __ Ret(pl);  // Return smi 0 if the non-zero one was positive.
+      __ add(scratch2, right, left);
+      __ cmpge(scratch2, Operand(0));
+      __ bf_near(&skip_if_false);
+      __ mov(right, Operand(Smi::FromInt(0)));
+      __ rts();  // Return smi 0 if the non-zero one was positive.
+      __ bind(&skip_if_false);
       // We fall through here if we multiplied a negative number with 0, because
       // that would mean we should produce -0.
       break;
     case Token::DIV:
       // Check for power of two on the right hand side.
       __ JumpIfNotPowerOfTwoOrZero(right, scratch1, &not_smi_result);
       // Check for positive and no remainder (scratch1 contains right - 1).
       __ orr(scratch2, scratch1, Operand(0x80000000u));
       __ tst(left, scratch2);
       __ b(ne, &not_smi_result);
 
       // Perform division by shifting.
       __ CountLeadingZeros(scratch1, scratch1, scratch2);
       __ rsb(scratch1, scratch1, Operand(31));
-      __ mov(right, Operand(left, LSR, scratch1));
+      __ lsr(right, left, scratch1);
       __ Ret();
       break;
     case Token::MOD:
       // Check for two positive smis.
-      __ orr(scratch1, left, Operand(right));
+      __ orr(scratch1, left, right);
       __ tst(scratch1, Operand(0x80000000u | kSmiTagMask));
       __ b(ne, &not_smi_result);
 
       // Check for power of two on the right hand side.
       __ JumpIfNotPowerOfTwoOrZero(right, scratch1, &not_smi_result);
 
       // Perform modulus by masking.
-      __ and_(right, left, Operand(scratch1));
+      __ land(right, left, scratch1);
       __ Ret();
       break;
     case Token::BIT_OR:
-      __ orr(right, left, Operand(right));
+      __ orr(right, left, right);
       __ Ret();
       break;
     case Token::BIT_AND:
-      __ and_(right, left, Operand(right));
+      __ land(right, left, right);
       __ Ret();
       break;
     case Token::BIT_XOR:
-      __ eor(right, left, Operand(right));
+      __ eor(right, left, right);
       __ Ret();
       break;
     case Token::SAR:
       // Remove tags from right operand.
       __ GetLeastBitsFromSmi(scratch1, right, 5);
-      __ mov(right, Operand(left, ASR, scratch1));
+      __ asr(right, left, scratch1);
       // Smi tag result.
       __ bic(right, right, Operand(kSmiTagMask));
       __ Ret();
       break;
     case Token::SHR:
       // Remove tags from operands. We can't do this on a 31 bit number
       // because then the 0s get shifted into bit 30 instead of bit 31.
       __ SmiUntag(scratch1, left);
       __ GetLeastBitsFromSmi(scratch2, right, 5);
-      __ mov(scratch1, Operand(scratch1, LSR, scratch2));
+      __ lsr(scratch1, scratch1, scratch2);
       // Unsigned shift is not allowed to produce a negative number, so
       // check the sign bit and the sign bit after Smi tagging.
       __ tst(scratch1, Operand(0xc0000000));
       __ b(ne, &not_smi_result);
       // Smi tag result.
       __ SmiTag(right, scratch1);
       __ Ret();
       break;
     case Token::SHL:
       // Remove tags from operands.
       __ SmiUntag(scratch1, left);
       __ GetLeastBitsFromSmi(scratch2, right, 5);
-      __ mov(scratch1, Operand(scratch1, LSL, scratch2));
+      __ lsl(scratch1, scratch1, scratch2);
       // Check that the signed result fits in a Smi.
-      __ add(scratch2, scratch1, Operand(0x40000000), SetCC);
-      __ b(mi, &not_smi_result);
+      __ add(scratch2, scratch1, Operand(0x40000000));
+      __ cmpge(scratch2, Operand(0));
+      __ bf(&not_smi_result);
       __ SmiTag(right, scratch1);
       __ Ret();
       break;
     default:
       UNREACHABLE();
   }
   __ bind(&not_smi_result);
 }
 
 
 void BinaryOpStub::GenerateFPOperation(MacroAssembler* masm,
                                        bool smi_operands,
                                        Label* not_numbers,
                                        Label* gc_required) {
   Register left = r1;
   Register right = r0;
   Register scratch1 = r7;
   Register scratch2 = r9;
   Register scratch3 = r4;
 
   ASSERT(smi_operands || (not_numbers != NULL));
-  if (smi_operands) {
-    __ AssertSmi(left);
-    __ AssertSmi(right);
+  if (smi_operands && FLAG_debug_code) {
+    __ AbortIfNotSmi(left);
+    __ AbortIfNotSmi(right);
   }
 
   Register heap_number_map = r6;
   __ LoadRoot(heap_number_map, Heap::kHeapNumberMapRootIndex);
 
   switch (op_) {
     case Token::ADD:
     case Token::SUB:
     case Token::MUL:
     case Token::DIV:
     case Token::MOD: {
       // Load left and right operands into d6 and d7 or r0/r1 and r2/r3
       // depending on whether VFP3 is available or not.
       FloatingPointHelper::Destination destination =
-          CpuFeatures::IsSupported(VFP2) &&
+          CpuFeatures::IsSupported(FPU) &&
           op_ != Token::MOD ?
           FloatingPointHelper::kVFPRegisters :
           FloatingPointHelper::kCoreRegisters;
 
       // Allocate new heap number for result.
       Register result = r5;
       GenerateHeapResultAllocation(
           masm, result, heap_number_map, scratch1, scratch2, gc_required);
 
       // Load the operands.
       if (smi_operands) {
         FloatingPointHelper::LoadSmis(masm, destination, scratch1, scratch2);
       } else {
         FloatingPointHelper::LoadOperands(masm,
                                           destination,
                                           heap_number_map,
                                           scratch1,
                                           scratch2,
                                           not_numbers);
       }
 
       // Calculate the result.
       if (destination == FloatingPointHelper::kVFPRegisters) {
-        // Using VFP registers:
-        // d6: Left value
-        // d7: Right value
-        CpuFeatures::Scope scope(VFP2);
+        // Using FPU registers:
+        // dr0: Left value
+        // dr2: Right value
         switch (op_) {
           case Token::ADD:
-            __ vadd(d5, d6, d7);
+            __ fadd(dr0, dr2);
             break;
           case Token::SUB:
-            __ vsub(d5, d6, d7);
+            __ fsub(dr0, dr2);
             break;
           case Token::MUL:
-            __ vmul(d5, d6, d7);
+            __ fmul(dr0, dr2);
             break;
           case Token::DIV:
-            __ vdiv(d5, d6, d7);
+            __ fdiv(dr0, dr2);
             break;
           default:
             UNREACHABLE();
         }
 
-        __ sub(r0, result, Operand(kHeapObjectTag));
-        __ vstr(d5, r0, HeapNumber::kValueOffset);
-        __ add(r0, r0, Operand(kHeapObjectTag));
+        __ sub(r0, result, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+        __ dstr(dr0, MemOperand(r0, 0));
+        __ add(r0, r0, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
         __ Ret();
       } else {
         // Call the C function to handle the double operation.
         FloatingPointHelper::CallCCodeForDoubleOperation(masm,
                                                          op_,
                                                          result,
                                                          scratch1);
         if (FLAG_debug_code) {
           __ stop("Unreachable code.");
         }
@@ skipping 11 matching lines @@
         __ SmiUntag(r2, right);
       } else {
         // Convert operands to 32-bit integers. Right in r2 and left in r3.
         FloatingPointHelper::ConvertNumberToInt32(masm,
                                                   left,
                                                   r3,
                                                   heap_number_map,
                                                   scratch1,
                                                   scratch2,
                                                   scratch3,
-                                                  d0,
+                                                  dr0,
                                                   not_numbers);
         FloatingPointHelper::ConvertNumberToInt32(masm,
                                                   right,
                                                   r2,
                                                   heap_number_map,
                                                   scratch1,
                                                   scratch2,
                                                   scratch3,
-                                                  d0,
+                                                  dr0,
                                                   not_numbers);
       }
 
       Label result_not_a_smi;
       switch (op_) {
         case Token::BIT_OR:
-          __ orr(r2, r3, Operand(r2));
+          __ orr(r2, r3, r2);
           break;
         case Token::BIT_XOR:
-          __ eor(r2, r3, Operand(r2));
+          __ eor(r2, r3, r2);
           break;
         case Token::BIT_AND:
-          __ and_(r2, r3, Operand(r2));
+          __ land(r2, r3, r2);
           break;
         case Token::SAR:
           // Use only the 5 least significant bits of the shift count.
           __ GetLeastBitsFromInt32(r2, r2, 5);
-          __ mov(r2, Operand(r3, ASR, r2));
+          __ asr(r2, r3, r2);
           break;
         case Token::SHR:
           // Use only the 5 least significant bits of the shift count.
           __ GetLeastBitsFromInt32(r2, r2, 5);
-          __ mov(r2, Operand(r3, LSR, r2), SetCC);
+          __ lsr(r2, r3, r2);
+          __ cmpge(r2, Operand(0));  // Check non-negative (see comment below).
           // SHR is special because it is required to produce a positive answer.
           // The code below for writing into heap numbers isn't capable of
           // writing the register as an unsigned int so we go to slow case if we
           // hit this case.
-          if (CpuFeatures::IsSupported(VFP2)) {
-            __ b(mi, &result_not_a_smi);
+          if (CpuFeatures::IsSupported(FPU)) {
+            __ bf(&result_not_a_smi);
           } else {
-            __ b(mi, not_numbers);
+            __ bf(not_numbers);
           }
           break;
         case Token::SHL:
           // Use only the 5 least significant bits of the shift count.
           __ GetLeastBitsFromInt32(r2, r2, 5);
-          __ mov(r2, Operand(r3, LSL, r2));
+          __ lsl(r2, r3, r2);
           break;
         default:
           UNREACHABLE();
       }
 
       // Check that the *signed* result fits in a smi.
-      __ add(r3, r2, Operand(0x40000000), SetCC);
-      __ b(mi, &result_not_a_smi);
+      __ add(r3, r2, Operand(0x40000000));
+      __ cmpge(r3, Operand(0));
+      __ bf(&result_not_a_smi);
       __ SmiTag(r0, r2);
       __ Ret();
 
       // Allocate new heap number for result.
       __ bind(&result_not_a_smi);
       Register result = r5;
       if (smi_operands) {
         __ AllocateHeapNumber(
             result, scratch1, scratch2, heap_number_map, gc_required);
       } else {
         GenerateHeapResultAllocation(
             masm, result, heap_number_map, scratch1, scratch2, gc_required);
       }
 
       // r2: Answer as signed int32.
       // r5: Heap number to write answer into.
 
       // Nothing can go wrong now, so move the heap number to r0, which is the
       // result.
-      __ mov(r0, Operand(r5));
+      __ mov(r0, r5);       // TODO(stm): look at this: it should be better
 
-      if (CpuFeatures::IsSupported(VFP2)) {
-        // Convert the int32 in r2 to the heap number in r0. r3 is corrupted. As
-        // mentioned above SHR needs to always produce a positive result.
-        CpuFeatures::Scope scope(VFP2);
-        __ vmov(s0, r2);
+      if (CpuFeatures::IsSupported(FPU)) {
         if (op_ == Token::SHR) {
-          __ vcvt_f64_u32(d0, s0);
+            __ dufloat(dr0, r2, dr2, sh4_rtmp);
         } else {
-          __ vcvt_f64_s32(d0, s0);
+          __ dfloat(dr0, r2);
         }
-        __ sub(r3, r0, Operand(kHeapObjectTag));
-        __ vstr(d0, r3, HeapNumber::kValueOffset);
+        __ sub(r3, r0, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+        __ dstr(dr0, MemOperand(r3, 0), r3);
         __ Ret();
       } else {
         // Tail call that writes the int32 in r2 to the heap number in r0, using
         // r3 as scratch. r0 is preserved and returned.
         WriteInt32ToHeapNumberStub stub(r2, r0, r3);
         __ TailCallStub(&stub);
       }
       break;
     }
     default:
@@ skipping 11 matching lines @@
     Label* use_runtime,
     Label* gc_required,
     SmiCodeGenerateHeapNumberResults allow_heapnumber_results) {
   Label not_smis;
 
   Register left = r1;
   Register right = r0;
   Register scratch1 = r7;
 
   // Perform combined smi check on both operands.
-  __ orr(scratch1, left, Operand(right));
+  __ orr(scratch1, left, right);
   STATIC_ASSERT(kSmiTag == 0);
   __ JumpIfNotSmi(scratch1, &not_smis);
 
   // If the smi-smi operation results in a smi return is generated.
   GenerateSmiSmiOperation(masm);
 
   // If heap number results are possible generate the result in an allocated
   // heap number.
   if (allow_heapnumber_results == ALLOW_HEAPNUMBER_RESULTS) {
     GenerateFPOperation(masm, true, use_runtime, gc_required);
@@ skipping 43 matching lines @@
   ASSERT(op_ == Token::ADD);
   // If both arguments are strings, call the string add stub.
   // Otherwise, do a transition.
 
   // Registers containing left and right operands respectively.
   Register left = r1;
   Register right = r0;
 
   // Test if left operand is a string.
   __ JumpIfSmi(left, &call_runtime);
-  __ CompareObjectType(left, r2, r2, FIRST_NONSTRING_TYPE);
-  __ b(ge, &call_runtime);
+  __ CompareObjectType(left, r2, r2, FIRST_NONSTRING_TYPE, ge);
+  __ bt(&call_runtime);
 
   // Test if right operand is a string.
   __ JumpIfSmi(right, &call_runtime);
-  __ CompareObjectType(right, r2, r2, FIRST_NONSTRING_TYPE);
-  __ b(ge, &call_runtime);
+  __ CompareObjectType(right, r2, r2, FIRST_NONSTRING_TYPE, ge);
+  __ bt(&call_runtime);
 
   StringAddStub string_add_stub(NO_STRING_CHECK_IN_STUB);
   GenerateRegisterArgsPush(masm);
   __ TailCallStub(&string_add_stub);
 
   __ bind(&call_runtime);
   GenerateTypeTransition(masm);
 }
 
 
 void BinaryOpStub::GenerateInt32Stub(MacroAssembler* masm) {
   ASSERT(operands_type_ == BinaryOpIC::INT32);
 
   Register left = r1;
   Register right = r0;
   Register scratch1 = r7;
   Register scratch2 = r9;
-  DwVfpRegister double_scratch = d0;
+  DwVfpRegister double_scratch = dr0;
 
   Register heap_number_result = no_reg;
   Register heap_number_map = r6;
   __ LoadRoot(heap_number_map, Heap::kHeapNumberMapRootIndex);
 
   Label call_runtime;
   // Labels for type transition, used for wrong input or output types.
   // Both label are currently actually bound to the same position. We use two
   // different label to differentiate the cause leading to type transition.
   Label transition;
 
   // Smi-smi fast case.
   Label skip;
   __ orr(scratch1, left, right);
   __ JumpIfNotSmi(scratch1, &skip);
   GenerateSmiSmiOperation(masm);
   // Fall through if the result is not a smi.
   __ bind(&skip);
 
   switch (op_) {
     case Token::ADD:
     case Token::SUB:
     case Token::MUL:
     case Token::DIV:
     case Token::MOD: {
       // Load both operands and check that they are 32-bit integer.
       // Jump to type transition if they are not. The registers r0 and r1 (right
       // and left) are preserved for the runtime call.
       FloatingPointHelper::Destination destination =
-          (CpuFeatures::IsSupported(VFP2) && op_ != Token::MOD)
+          (CpuFeatures::IsSupported(FPU) && op_ != Token::MOD)
               ? FloatingPointHelper::kVFPRegisters
               : FloatingPointHelper::kCoreRegisters;
 
       FloatingPointHelper::LoadNumberAsInt32Double(masm,
                                                    right,
                                                    destination,
-                                                   d7,
-                                                   d8,
+                                                   dr2,
                                                    r2,
                                                    r3,
                                                    heap_number_map,
                                                    scratch1,
                                                    scratch2,
-                                                   s0,
+                                                   fr4,
                                                    &transition);
       FloatingPointHelper::LoadNumberAsInt32Double(masm,
                                                    left,
                                                    destination,
-                                                   d6,
-                                                   d8,
+                                                   dr0,
                                                    r4,
                                                    r5,
                                                    heap_number_map,
                                                    scratch1,
                                                    scratch2,
-                                                   s0,
+                                                   fr4,
                                                    &transition);
 
-      if (destination == FloatingPointHelper::kVFPRegisters) {
-        CpuFeatures::Scope scope(VFP2);
+    if (destination == FloatingPointHelper::kVFPRegisters) {
         Label return_heap_number;
         switch (op_) {
           case Token::ADD:
-            __ vadd(d5, d6, d7);
+            __ fadd(dr0, dr2);
             break;
           case Token::SUB:
-            __ vsub(d5, d6, d7);
+            __ fsub(dr0, dr2);
             break;
           case Token::MUL:
-            __ vmul(d5, d6, d7);
+            __ fmul(dr0, dr2);
             break;
           case Token::DIV:
-            __ vdiv(d5, d6, d7);
+            __ fdiv(dr0, dr2);
             break;
           default:
             UNREACHABLE();
         }
 
         if (op_ != Token::DIV) {
           // These operations produce an integer result.
           // Try to return a smi if we can.
           // Otherwise return a heap number if allowed, or jump to type
           // transition.
 
-          __ EmitVFPTruncate(kRoundToZero,
-                             scratch1,
-                             d5,
+          __ EmitFPUTruncate(kRoundToZero,
                              scratch2,
-                             d8);
+                             dr0,
+                             scratch1);
 
           if (result_type_ <= BinaryOpIC::INT32) {
             // If the ne condition is set, result does
             // not fit in a 32-bit integer.
             __ b(ne, &transition);
           }
 
           // Check if the result fits in a smi.
-          __ add(scratch2, scratch1, Operand(0x40000000), SetCC);
+          __ add(scratch2, scratch1, Operand(0x40000000));
+          __ cmpge(scratch2, Operand(0));
           // If not try to return a heap number.
-          __ b(mi, &return_heap_number);
+          __ bt(&return_heap_number);
           // Check for minus zero. Return heap number for minus zero.
           Label not_zero;
-          __ cmp(scratch1, Operand::Zero());
+          __ cmp(scratch1, Operand(0));
           __ b(ne, &not_zero);
-          __ vmov(scratch2, d5.high());
+          __ isingle(scratch2, dr0.high());
           __ tst(scratch2, Operand(HeapNumber::kSignMask));
           __ b(ne, &return_heap_number);
           __ bind(&not_zero);
 
           // Tag the result and return.
           __ SmiTag(r0, scratch1);
           __ Ret();
         } else {
           // DIV just falls through to allocating a heap number.
         }
 
         __ bind(&return_heap_number);
         // Return a heap number, or fall through to type transition or runtime
         // call if we can't.
         if (result_type_ >= ((op_ == Token::DIV) ? BinaryOpIC::HEAP_NUMBER
                                                  : BinaryOpIC::INT32)) {
           // We are using vfp registers so r5 is available.
           heap_number_result = r5;
           GenerateHeapResultAllocation(masm,
                                        heap_number_result,
                                        heap_number_map,
                                        scratch1,
                                        scratch2,
                                        &call_runtime);
           __ sub(r0, heap_number_result, Operand(kHeapObjectTag));
-          __ vstr(d5, r0, HeapNumber::kValueOffset);
+          __ dstr(dr0, MemOperand(r0, HeapNumber::kValueOffset));
           __ mov(r0, heap_number_result);
           __ Ret();
         }
 
         // A DIV operation expecting an integer result falls through
         // to type transition.
 
       } else {
         // We preserved r0 and r1 to be able to call runtime.
         // Save the left value on the stack.
@@ skipping 39 matching lines @@
       // Convert operands to 32-bit integers. Right in r2 and left in r3. The
       // registers r0 and r1 (right and left) are preserved for the runtime
       // call.
       FloatingPointHelper::LoadNumberAsInt32(masm,
                                              left,
                                              r3,
                                              heap_number_map,
                                              scratch1,
                                              scratch2,
                                              scratch3,
-                                             d0,
-                                             d1,
+                                             dr0,
                                              &transition);
       FloatingPointHelper::LoadNumberAsInt32(masm,
                                              right,
                                              r2,
                                              heap_number_map,
                                              scratch1,
                                              scratch2,
                                              scratch3,
-                                             d0,
-                                             d1,
+                                             dr0,
                                              &transition);
 
       // The ECMA-262 standard specifies that, for shift operations, only the
       // 5 least significant bits of the shift value should be used.
       switch (op_) {
         case Token::BIT_OR:
-          __ orr(r2, r3, Operand(r2));
+          __ orr(r2, r3, r2);
           break;
         case Token::BIT_XOR:
-          __ eor(r2, r3, Operand(r2));
+          __ eor(r2, r3, r2);
           break;
         case Token::BIT_AND:
-          __ and_(r2, r3, Operand(r2));
+          __ land(r2, r3, r2);
           break;
         case Token::SAR:
-          __ and_(r2, r2, Operand(0x1f));
-          __ mov(r2, Operand(r3, ASR, r2));
+          __ land(r2, r2, Operand(0x1f));
+          __ asr(r2, r3, r2);
           break;
         case Token::SHR:
-          __ and_(r2, r2, Operand(0x1f));
-          __ mov(r2, Operand(r3, LSR, r2), SetCC);
+          __ land(r2, r2, Operand(0x1f));
+          __ lsr(r2, r3, r2);
           // SHR is special because it is required to produce a positive answer.
           // We only get a negative result if the shift value (r2) is 0.
           // This result cannot be respresented as a signed 32-bit integer, try
           // to return a heap number if we can.
-          // The non vfp2 code does not support this special case, so jump to
+          // The non vfp3 code does not support this special case, so jump to
           // runtime if we don't support it.
-          if (CpuFeatures::IsSupported(VFP2)) {
-            __ b(mi, (result_type_ <= BinaryOpIC::INT32)
+          __ cmpge(r2, Operand(0));
+          if (CpuFeatures::IsSupported(FPU)) {
+            __ b(f, (result_type_ <= BinaryOpIC::INT32)
                       ? &transition
                       : &return_heap_number);
           } else {
-            __ b(mi, (result_type_ <= BinaryOpIC::INT32)
+            __ b(f, (result_type_ <= BinaryOpIC::INT32)
                       ? &transition
                       : &call_runtime);
           }
           break;
         case Token::SHL:
-          __ and_(r2, r2, Operand(0x1f));
-          __ mov(r2, Operand(r3, LSL, r2));
+          __ land(r2, r2, Operand(0x1f));
+          __ lsl(r2, r3, r2);
           break;
         default:
           UNREACHABLE();
       }
 
       // Check if the result fits in a smi.
-      __ add(scratch1, r2, Operand(0x40000000), SetCC);
+      __ add(scratch1, r2, Operand(0x40000000));
       // If not try to return a heap number. (We know the result is an int32.)
-      __ b(mi, &return_heap_number);
+      __ cmpge(scratch1, Operand(0));
+      __ b(f, &return_heap_number);
       // Tag the result and return.
       __ SmiTag(r0, r2);
       __ Ret();
 
       __ bind(&return_heap_number);
       heap_number_result = r5;
       GenerateHeapResultAllocation(masm,
                                    heap_number_result,
                                    heap_number_map,
                                    scratch1,
                                    scratch2,
                                    &call_runtime);
 
-      if (CpuFeatures::IsSupported(VFP2)) {
-        CpuFeatures::Scope scope(VFP2);
-        if (op_ != Token::SHR) {
+      if (CpuFeatures::IsSupported(FPU)) {
+        if ((op_ != Token::SHR)) {
           // Convert the result to a floating point value.
-          __ vmov(double_scratch.low(), r2);
-          __ vcvt_f64_s32(double_scratch, double_scratch.low());
+          __ dfloat(double_scratch, r2);
         } else {
-          // The result must be interpreted as an unsigned 32-bit integer.
-          __ vmov(double_scratch.low(), r2);
-          __ vcvt_f64_u32(double_scratch, double_scratch.low());
+          __ dufloat(double_scratch, r2, dr2, sh4_rtmp);
         }
 
         // Store the result.
         __ sub(r0, heap_number_result, Operand(kHeapObjectTag));
-        __ vstr(double_scratch, r0, HeapNumber::kValueOffset);
+        __ dstr(double_scratch, MemOperand(r0, HeapNumber::kValueOffset));
         __ mov(r0, heap_number_result);
         __ Ret();
       } else {
         // Tail call that writes the int32 in r2 to the heap number in r0, using
         // r3 as scratch. r0 is preserved and returned.
         __ mov(r0, r5);
         WriteInt32ToHeapNumberStub stub(r2, r0, r3);
         __ TailCallStub(&stub);
       }
 
@@ skipping 29 matching lines @@
 
   // Convert oddball arguments to numbers.
   Label check, done;
   __ CompareRoot(r1, Heap::kUndefinedValueRootIndex);
   __ b(ne, &check);
   if (Token::IsBitOp(op_)) {
     __ mov(r1, Operand(Smi::FromInt(0)));
   } else {
     __ LoadRoot(r1, Heap::kNanValueRootIndex);
   }
-  __ jmp(&done);
+  __ jmp_near(&done);
   __ bind(&check);
   __ CompareRoot(r0, Heap::kUndefinedValueRootIndex);
-  __ b(ne, &done);
+  __ b(ne, &done, Label::kNear);
   if (Token::IsBitOp(op_)) {
     __ mov(r0, Operand(Smi::FromInt(0)));
   } else {
     __ LoadRoot(r0, Heap::kNanValueRootIndex);
   }
   __ bind(&done);
 
   GenerateHeapNumberStub(masm);
 }
 
@@ skipping 26 matching lines @@
 
 void BinaryOpStub::GenerateAddStrings(MacroAssembler* masm) {
   ASSERT(op_ == Token::ADD);
   Label left_not_string, call_runtime;
 
   Register left = r1;
   Register right = r0;
 
   // Check if left argument is a string.
   __ JumpIfSmi(left, &left_not_string);
-  __ CompareObjectType(left, r2, r2, FIRST_NONSTRING_TYPE);
-  __ b(ge, &left_not_string);
+  __ CompareObjectType(left, r2, r2, FIRST_NONSTRING_TYPE, ge);
+  __ bt(&left_not_string);
 
   StringAddStub string_add_left_stub(NO_STRING_CHECK_LEFT_IN_STUB);
   GenerateRegisterArgsPush(masm);
   __ TailCallStub(&string_add_left_stub);
 
   // Left operand is not a string, test right.
   __ bind(&left_not_string);
   __ JumpIfSmi(right, &call_runtime);
-  __ CompareObjectType(right, r2, r2, FIRST_NONSTRING_TYPE);
-  __ b(ge, &call_runtime);
+  __ CompareObjectType(right, r2, r2, FIRST_NONSTRING_TYPE, ge);
+  __ bt(&call_runtime);
 
   StringAddStub string_add_right_stub(NO_STRING_CHECK_RIGHT_IN_STUB);
   GenerateRegisterArgsPush(masm);
   __ TailCallStub(&string_add_right_stub);
 
   // At least one argument is not a string.
   __ bind(&call_runtime);
 }
 
 
@@ skipping 54 matching lines @@
     Register overwritable_operand = mode_ == OVERWRITE_LEFT ? r1 : r0;
     // If the overwritable operand is already an object, we skip the
     // allocation of a heap number.
     __ JumpIfNotSmi(overwritable_operand, &skip_allocation);
     // Allocate a heap number for the result.
     __ AllocateHeapNumber(
         result, scratch1, scratch2, heap_number_map, gc_required);
     __ b(&allocated);
     __ bind(&skip_allocation);
     // Use object holding the overwritable operand for result.
-    __ mov(result, Operand(overwritable_operand));
+    __ mov(result, overwritable_operand);
     __ bind(&allocated);
   } else {
     ASSERT(mode_ == NO_OVERWRITE);
     __ AllocateHeapNumber(
         result, scratch1, scratch2, heap_number_map, gc_required);
   }
 }
 
 
 void BinaryOpStub::GenerateRegisterArgsPush(MacroAssembler* masm) {
   __ Push(r1, r0);
 }
 
 
 void TranscendentalCacheStub::Generate(MacroAssembler* masm) {
-  // Untagged case: double input in d2, double result goes
-  //   into d2.
+  // Untagged case: double input in dr2, double result goes
+  //   into dr2.
   // Tagged case: tagged input on top of stack and in r0,
   //   tagged result (heap number) goes into r0.
 
   Label input_not_smi;
   Label loaded;
   Label calculate;
   Label invalid_cache;
   const Register scratch0 = r9;
   const Register scratch1 = r7;
   const Register cache_entry = r0;
   const bool tagged = (argument_type_ == TAGGED);
 
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
+  if (CpuFeatures::IsSupported(FPU)) {
     if (tagged) {
       // Argument is a number and is on stack and in r0.
       // Load argument and check if it is a smi.
       __ JumpIfNotSmi(r0, &input_not_smi);
 
       // Input is a smi. Convert to double and load the low and high words
       // of the double into r2, r3.
-      __ IntegerToDoubleConversionWithVFP3(r0, r3, r2);
+      __ asr(scratch0, r0, Operand(kSmiTagSize));
+      __ dfloat(dr0, scratch0);
+      __ movd(r2, r3, dr0);
       __ b(&loaded);
 
       __ bind(&input_not_smi);
       // Check if input is a HeapNumber.
       __ CheckMap(r0,
                   r1,
                   Heap::kHeapNumberMapRootIndex,
                   &calculate,
                   DONT_DO_SMI_CHECK);
       // Input is a HeapNumber. Load it to a double register and store the
       // low and high words into r2, r3.
-      __ vldr(d0, FieldMemOperand(r0, HeapNumber::kValueOffset));
-      __ vmov(r2, r3, d0);
+      __ dldr(dr0, FieldMemOperand(r0, HeapNumber::kValueOffset));
+      __ movd(r2, r3, dr0);
     } else {
-      // Input is untagged double in d2. Output goes to d2.
-      __ vmov(r2, r3, d2);
+      UNIMPLEMENTED();
     }
     __ bind(&loaded);
     // r2 = low 32 bits of double value
     // r3 = high 32 bits of double value
     // Compute hash (the shifts are arithmetic):
     //   h = (low ^ high); h ^= h >> 16; h ^= h >> 8; h = h & (cacheSize - 1);
-    __ eor(r1, r2, Operand(r3));
-    __ eor(r1, r1, Operand(r1, ASR, 16));
-    __ eor(r1, r1, Operand(r1, ASR, 8));
+    __ eor(r1, r2, r3);
+    __ asr(scratch0, r1, Operand(16));
+    __ eor(r1, r1, scratch0);
+    __ asr(scratch0, r1, Operand(8));
+    __ eor(r1, r1, scratch0);
     ASSERT(IsPowerOf2(TranscendentalCache::SubCache::kCacheSize));
-    __ And(r1, r1, Operand(TranscendentalCache::SubCache::kCacheSize - 1));
+    // TODO(STM): ??
+    __ land(r1, r1, Operand(TranscendentalCache::SubCache::kCacheSize - 1));
 
     // r2 = low 32 bits of double value.
     // r3 = high 32 bits of double value.
     // r1 = TranscendentalCache::hash(double value).
     Isolate* isolate = masm->isolate();
     ExternalReference cache_array =
         ExternalReference::transcendental_cache_array_address(isolate);
     __ mov(cache_entry, Operand(cache_array));
     // cache_entry points to cache array.
     int cache_array_index
@@ skipping 13 matching lines @@
       char* elem_in1 = reinterpret_cast<char*>(&(test_elem[0].in[1]));
       char* elem_out = reinterpret_cast<char*>(&(test_elem[0].output));
       CHECK_EQ(12, elem2_start - elem_start);  // Two uint_32's and a pointer.
       CHECK_EQ(0, elem_in0 - elem_start);
       CHECK_EQ(kIntSize, elem_in1 - elem_start);
       CHECK_EQ(2 * kIntSize, elem_out - elem_start);
     }
 #endif
 
     // Find the address of the r1'st entry in the cache, i.e., &r0[r1*12].
-    __ add(r1, r1, Operand(r1, LSL, 1));
-    __ add(cache_entry, cache_entry, Operand(r1, LSL, 2));
+    __ lsl(scratch0, r1, Operand(1));
+    __ add(r1, r1, scratch0);
+    __ lsl(scratch0, r1, Operand(2));
+    __ add(cache_entry, cache_entry, scratch0);
     // Check if cache matches: Double value is stored in uint32_t[2] array.
-    __ ldm(ia, cache_entry, r4.bit() | r5.bit() | r6.bit());
+    __ ldr(r4, MemOperand(cache_entry, 0));
+    __ ldr(r5, MemOperand(cache_entry, 4));
+    __ ldr(r6, MemOperand(cache_entry, 8));
     __ cmp(r2, r4);
-    __ cmp(r3, r5, eq);
+    __ b(ne, &calculate);
+    __ cmp(r3, r5);
     __ b(ne, &calculate);
     // Cache hit. Load result, cleanup and return.
     Counters* counters = masm->isolate()->counters();
     __ IncrementCounter(
         counters->transcendental_cache_hit(), 1, scratch0, scratch1);
     if (tagged) {
       // Pop input value from stack and load result into r0.
       __ pop();
-      __ mov(r0, Operand(r6));
+      __ mov(r0, r6);
     } else {
-      // Load result into d2.
-       __ vldr(d2, FieldMemOperand(r6, HeapNumber::kValueOffset));
+      // Load result into dr2.
+       __ dldr(dr2, FieldMemOperand(r6, HeapNumber::kValueOffset));
     }
     __ Ret();
-  }  // if (CpuFeatures::IsSupported(VFP3))
+  }  // if (CpuFeatures::IsSupported(FPU))
 
   __ bind(&calculate);
   Counters* counters = masm->isolate()->counters();
   __ IncrementCounter(
       counters->transcendental_cache_miss(), 1, scratch0, scratch1);
   if (tagged) {
     __ bind(&invalid_cache);
     ExternalReference runtime_function =
         ExternalReference(RuntimeFunction(), masm->isolate());
     __ TailCallExternalReference(runtime_function, 1, 1);
   } else {
-    ASSERT(CpuFeatures::IsSupported(VFP2));
-    CpuFeatures::Scope scope(VFP2);
-
-    Label no_update;
-    Label skip_cache;
-
-    // Call C function to calculate the result and update the cache.
-    // r0: precalculated cache entry address.
-    // r2 and r3: parts of the double value.
-    // Store r0, r2 and r3 on stack for later before calling C function.
-    __ Push(r3, r2, cache_entry);
-    GenerateCallCFunction(masm, scratch0);
-    __ GetCFunctionDoubleResult(d2);
-
-    // Try to update the cache. If we cannot allocate a
-    // heap number, we return the result without updating.
-    __ Pop(r3, r2, cache_entry);
-    __ LoadRoot(r5, Heap::kHeapNumberMapRootIndex);
-    __ AllocateHeapNumber(r6, scratch0, scratch1, r5, &no_update);
-    __ vstr(d2, FieldMemOperand(r6, HeapNumber::kValueOffset));
-    __ stm(ia, cache_entry, r2.bit() | r3.bit() | r6.bit());
-    __ Ret();
-
-    __ bind(&invalid_cache);
-    // The cache is invalid. Call runtime which will recreate the
-    // cache.
-    __ LoadRoot(r5, Heap::kHeapNumberMapRootIndex);
-    __ AllocateHeapNumber(r0, scratch0, scratch1, r5, &skip_cache);
-    __ vstr(d2, FieldMemOperand(r0, HeapNumber::kValueOffset));
-    {
-      FrameScope scope(masm, StackFrame::INTERNAL);
-      __ push(r0);
-      __ CallRuntime(RuntimeFunction(), 1);
-    }
-    __ vldr(d2, FieldMemOperand(r0, HeapNumber::kValueOffset));
-    __ Ret();
-
-    __ bind(&skip_cache);
-    // Call C function to calculate the result and answer directly
-    // without updating the cache.
-    GenerateCallCFunction(masm, scratch0);
-    __ GetCFunctionDoubleResult(d2);
-    __ bind(&no_update);
-
-    // We return the value in d2 without adding it to the cache, but
-    // we cause a scavenging GC so that future allocations will succeed.
-    {
-      FrameScope scope(masm, StackFrame::INTERNAL);
-
-      // Allocate an aligned object larger than a HeapNumber.
-      ASSERT(4 * kPointerSize >= HeapNumber::kSize);
-      __ mov(scratch0, Operand(4 * kPointerSize));
-      __ push(scratch0);
-      __ CallRuntimeSaveDoubles(Runtime::kAllocateInNewSpace);
-    }
-    __ Ret();
+    UNREACHABLE();
   }
 }
 
 
 void TranscendentalCacheStub::GenerateCallCFunction(MacroAssembler* masm,
                                                     Register scratch) {
-  ASSERT(CpuFeatures::IsEnabled(VFP2));
   Isolate* isolate = masm->isolate();
 
   __ push(lr);
   __ PrepareCallCFunction(0, 1, scratch);
-  if (masm->use_eabi_hardfloat()) {
-    __ vmov(d0, d2);
-  } else {
-    __ vmov(r0, r1, d2);
-  }
-  AllowExternalCallThatCantCauseGC scope(masm);
+  __ movd(dr4, r0, r1);
   switch (type_) {
     case TranscendentalCache::SIN:
       __ CallCFunction(ExternalReference::math_sin_double_function(isolate),
           0, 1);
       break;
     case TranscendentalCache::COS:
       __ CallCFunction(ExternalReference::math_cos_double_function(isolate),
           0, 1);
       break;
     case TranscendentalCache::TAN:
@@ skipping 30 matching lines @@
   __ TailCallRuntime(Runtime::kStackGuard, 0, 1);
 }
 
 
 void InterruptStub::Generate(MacroAssembler* masm) {
   __ TailCallRuntime(Runtime::kInterrupt, 0, 1);
 }
 
 
 void MathPowStub::Generate(MacroAssembler* masm) {
-  CpuFeatures::Scope vfp2_scope(VFP2);
-  const Register base = r1;
-  const Register exponent = r2;
-  const Register heapnumbermap = r5;
-  const Register heapnumber = r0;
-  const DoubleRegister double_base = d1;
-  const DoubleRegister double_exponent = d2;
-  const DoubleRegister double_result = d3;
-  const DoubleRegister double_scratch = d0;
-  const SwVfpRegister single_scratch = s0;
-  const Register scratch = r9;
-  const Register scratch2 = r7;
+  // TODO(STM): not merged !
+  Label call_runtime;
+  if (CpuFeatures::IsSupported(FPU)) {
+    Label base_not_smi;
+    Label exponent_not_smi;
+    Label convert_exponent;
 
-  Label call_runtime, done, int_exponent;
-  if (exponent_type_ == ON_STACK) {
-    Label base_is_smi, unpack_exponent;
-    // The exponent and base are supplied as arguments on the stack.
-    // This can only happen if the stub is called from non-optimized code.
-    // Load input parameters from stack to double registers.
+    const Register base = r0;
+    const Register exponent = r4;
+    const Register heapnumbermap = sh4_r8;
+    const Register heapnumber = r9;
+    const DoubleRegister double_base = dr4;
+    const DoubleRegister double_exponent = dr6;
+    const DoubleRegister double_result = dr0;
+    const Register scratch = r5;
+    const Register scratch2 = r6;
+
+    __ LoadRoot(heapnumbermap, Heap::kHeapNumberMapRootIndex);
     __ ldr(base, MemOperand(sp, 1 * kPointerSize));
     __ ldr(exponent, MemOperand(sp, 0 * kPointerSize));
 
-    __ LoadRoot(heapnumbermap, Heap::kHeapNumberMapRootIndex);
+    // Convert base to double value and store it in dr0.
+    __ JumpIfNotSmi(base, &base_not_smi);
+    // Base is a Smi. Untag and convert it.
+    __ SmiUntag(base);
+    __ dfloat(double_base, base);
+    __ b(&convert_exponent);
 
-    __ UntagAndJumpIfSmi(scratch, base, &base_is_smi);
+    __ bind(&base_not_smi);
     __ ldr(scratch, FieldMemOperand(base, JSObject::kMapOffset));
     __ cmp(scratch, heapnumbermap);
     __ b(ne, &call_runtime);
+    // Base is a heapnumber. Load it into double register.
+    __ dldr(double_base, FieldMemOperand(base, HeapNumber::kValueOffset));
 
-    __ vldr(double_base, FieldMemOperand(base, HeapNumber::kValueOffset));
-    __ jmp(&unpack_exponent);
+    __ bind(&convert_exponent);
+    __ JumpIfNotSmi(exponent, &exponent_not_smi);
+    __ SmiUntag(exponent);
 
-    __ bind(&base_is_smi);
-    __ vmov(single_scratch, scratch);
-    __ vcvt_f64_s32(double_base, single_scratch);
-    __ bind(&unpack_exponent);
+    // The base is in a double register and the exponent is
+    // an untagged smi. Allocate a heap number and call a
+    // C function for integer exponents. The register containing
+    // the heap number is callee-saved.
+    __ AllocateHeapNumber(heapnumber,
+                          scratch,
+                          scratch2,
+                          heapnumbermap,
+                          &call_runtime);
+    __ push(pr);
+    __ PrepareCallCFunction(1, 1, scratch);
+    // check that the argument are stored in the right registers (sh4 ABI)
+    ASSERT(double_base.is(dr4) && exponent.is(r4));
+    __ CallCFunction(
+        ExternalReference::power_double_int_function(masm->isolate()),
+        1, 1);
+    __ pop(pr);
+    ASSERT(double_result.is(dr0));
+    __ dstr(double_result,
+            FieldMemOperand(heapnumber, HeapNumber::kValueOffset));
+    __ mov(r0, heapnumber);
+    __ Drop(2);
+    __ rts();
 
-    __ UntagAndJumpIfSmi(scratch, exponent, &int_exponent);
-
+    __ bind(&exponent_not_smi);
     __ ldr(scratch, FieldMemOperand(exponent, JSObject::kMapOffset));
     __ cmp(scratch, heapnumbermap);
     __ b(ne, &call_runtime);
-    __ vldr(double_exponent,
+    // Exponent is a heapnumber. Load it into double register.
+    __ dldr(double_exponent,
             FieldMemOperand(exponent, HeapNumber::kValueOffset));
-  } else if (exponent_type_ == TAGGED) {
-    // Base is already in double_base.
-    __ UntagAndJumpIfSmi(scratch, exponent, &int_exponent);
 
-    __ vldr(double_exponent,
-            FieldMemOperand(exponent, HeapNumber::kValueOffset));
+    // The base and the exponent are in double registers.
+    // Allocate a heap number and call a C function for
+    // double exponents. The register containing
+    // the heap number is callee-saved.
+    __ AllocateHeapNumber(heapnumber,
+                          scratch,
+                          scratch2,
+                          heapnumbermap,
+                          &call_runtime);
+    __ push(pr);
+    __ PrepareCallCFunction(0, 2, scratch);
+    ASSERT(double_base.is(dr4) && double_exponent.is(dr6));
+    __ CallCFunction(
+        ExternalReference::power_double_double_function(masm->isolate()),
+        0, 2);
+    __ pop(pr);
+    ASSERT(double_result.is(dr0));
+    __ dstr(double_result,
+            FieldMemOperand(heapnumber, HeapNumber::kValueOffset));
+    __ mov(r0, heapnumber);
+    __ Drop(2);
+    __ rts();
   }
-
-  if (exponent_type_ != INTEGER) {
-    Label int_exponent_convert;
-    // Detect integer exponents stored as double.
-    __ vcvt_u32_f64(single_scratch, double_exponent);
-    // We do not check for NaN or Infinity here because comparing numbers on
-    // ARM correctly distinguishes NaNs.  We end up calling the built-in.
-    __ vcvt_f64_u32(double_scratch, single_scratch);
-    __ VFPCompareAndSetFlags(double_scratch, double_exponent);
-    __ b(eq, &int_exponent_convert);
-
-    if (exponent_type_ == ON_STACK) {
-      // Detect square root case.  Crankshaft detects constant +/-0.5 at
-      // compile time and uses DoMathPowHalf instead.  We then skip this check
-      // for non-constant cases of +/-0.5 as these hardly occur.
-      Label not_plus_half;
-
-      // Test for 0.5.
-      __ vmov(double_scratch, 0.5, scratch);
-      __ VFPCompareAndSetFlags(double_exponent, double_scratch);
-      __ b(ne, &not_plus_half);
-
-      // Calculates square root of base.  Check for the special case of
-      // Math.pow(-Infinity, 0.5) == Infinity (ECMA spec, 15.8.2.13).
-      __ vmov(double_scratch, -V8_INFINITY, scratch);
-      __ VFPCompareAndSetFlags(double_base, double_scratch);
-      __ vneg(double_result, double_scratch, eq);
-      __ b(eq, &done);
-
-      // Add +0 to convert -0 to +0.
-      __ vadd(double_scratch, double_base, kDoubleRegZero);
-      __ vsqrt(double_result, double_scratch);
-      __ jmp(&done);
-
-      __ bind(&not_plus_half);
-      __ vmov(double_scratch, -0.5, scratch);
-      __ VFPCompareAndSetFlags(double_exponent, double_scratch);
-      __ b(ne, &call_runtime);
-
-      // Calculates square root of base.  Check for the special case of
-      // Math.pow(-Infinity, -0.5) == 0 (ECMA spec, 15.8.2.13).
-      __ vmov(double_scratch, -V8_INFINITY, scratch);
-      __ VFPCompareAndSetFlags(double_base, double_scratch);
-      __ vmov(double_result, kDoubleRegZero, eq);
-      __ b(eq, &done);
-
-      // Add +0 to convert -0 to +0.
-      __ vadd(double_scratch, double_base, kDoubleRegZero);
-      __ vmov(double_result, 1.0, scratch);
-      __ vsqrt(double_scratch, double_scratch);
-      __ vdiv(double_result, double_result, double_scratch);
-      __ jmp(&done);
-    }
-
-    __ push(lr);
-    {
-      AllowExternalCallThatCantCauseGC scope(masm);
-      __ PrepareCallCFunction(0, 2, scratch);
-      __ SetCallCDoubleArguments(double_base, double_exponent);
-      __ CallCFunction(
-          ExternalReference::power_double_double_function(masm->isolate()),
-          0, 2);
-    }
-    __ pop(lr);
-    __ GetCFunctionDoubleResult(double_result);
-    __ jmp(&done);
-
-    __ bind(&int_exponent_convert);
-    __ vcvt_u32_f64(single_scratch, double_exponent);
-    __ vmov(scratch, single_scratch);
-  }
-
-  // Calculate power with integer exponent.
-  __ bind(&int_exponent);
-
-  // Get two copies of exponent in the registers scratch and exponent.
-  if (exponent_type_ == INTEGER) {
-    __ mov(scratch, exponent);
-  } else {
-    // Exponent has previously been stored into scratch as untagged integer.
-    __ mov(exponent, scratch);
-  }
-  __ vmov(double_scratch, double_base);  // Back up base.
-  __ vmov(double_result, 1.0, scratch2);
-
-  // Get absolute value of exponent.
-  __ cmp(scratch, Operand(0));
-  __ mov(scratch2, Operand(0), LeaveCC, mi);
-  __ sub(scratch, scratch2, scratch, LeaveCC, mi);
-
-  Label while_true;
-  __ bind(&while_true);
-  __ mov(scratch, Operand(scratch, ASR, 1), SetCC);
-  __ vmul(double_result, double_result, double_scratch, cs);
-  __ vmul(double_scratch, double_scratch, double_scratch, ne);
-  __ b(ne, &while_true);
-
-  __ cmp(exponent, Operand(0));
-  __ b(ge, &done);
-  __ vmov(double_scratch, 1.0, scratch);
-  __ vdiv(double_result, double_scratch, double_result);
-  // Test whether result is zero.  Bail out to check for subnormal result.
-  // Due to subnormals, x^-y == (1/x)^y does not hold in all cases.
-  __ VFPCompareAndSetFlags(double_result, 0.0);
-  __ b(ne, &done);
-  // double_exponent may not containe the exponent value if the input was a
-  // smi.  We set it with exponent value before bailing out.
-  __ vmov(single_scratch, exponent);
-  __ vcvt_f64_s32(double_exponent, single_scratch);
-
-  // Returning or bailing out.
-  Counters* counters = masm->isolate()->counters();
-  if (exponent_type_ == ON_STACK) {
-    // The arguments are still on the stack.
-    __ bind(&call_runtime);
-    __ TailCallRuntime(Runtime::kMath_pow_cfunction, 2, 1);
-
-    // The stub is called from non-optimized code, which expects the result
-    // as heap number in exponent.
-    __ bind(&done);
-    __ AllocateHeapNumber(
-        heapnumber, scratch, scratch2, heapnumbermap, &call_runtime);
-    __ vstr(double_result,
-            FieldMemOperand(heapnumber, HeapNumber::kValueOffset));
-    ASSERT(heapnumber.is(r0));
-    __ IncrementCounter(counters->math_pow(), 1, scratch, scratch2);
-    __ Ret(2);
-  } else {
-    __ push(lr);
-    {
-      AllowExternalCallThatCantCauseGC scope(masm);
-      __ PrepareCallCFunction(0, 2, scratch);
-      __ SetCallCDoubleArguments(double_base, double_exponent);
-      __ CallCFunction(
-          ExternalReference::power_double_double_function(masm->isolate()),
-          0, 2);
-    }
-    __ pop(lr);
-    __ GetCFunctionDoubleResult(double_result);
-
-    __ bind(&done);
-    __ IncrementCounter(counters->math_pow(), 1, scratch, scratch2);
-    __ Ret();
-  }
+  __ bind(&call_runtime);
+  __ TailCallRuntime(Runtime::kMath_pow_cfunction, 2, 1);
 }
 
 
 bool CEntryStub::NeedsImmovableCode() {
   return true;
 }
 
 
 bool CEntryStub::IsPregenerated() {
   return (!save_doubles_ || ISOLATE->fp_stubs_generated()) &&
@@ skipping 25 matching lines @@
   code->set_is_pregenerated(true);
 }
 
 
 void CEntryStub::GenerateCore(MacroAssembler* masm,
                               Label* throw_normal_exception,
                               Label* throw_termination_exception,
                               Label* throw_out_of_memory_exception,
                               bool do_gc,
                               bool always_allocate) {
+  // WARNING: this function use the SH4 ABI !!
+
+  // Input
   // r0: result parameter for PerformGC, if any
-  // r4: number of arguments including receiver  (C callee-saved)
-  // r5: pointer to builtin function  (C callee-saved)
-  // r6: pointer to the first argument (C callee-saved)
+  // sh4_r8: number of arguments including receiver  (C callee-saved)
+  //         Used later by LeaveExitFrame()
+  // sh4_r9: pointer to builtin function  (C callee-saved)
+  // sh4_r10: pointer to the first argument (C callee-saved)
+  // TODO(stm): use of r10 is dangerous (ip)
+  // sh4: moved callee-saved to stack localtion (see ::Generate())
   Isolate* isolate = masm->isolate();
+  ASSERT(!r0.is(sh4_rtmp));
+  ASSERT(!r0.is(sh4_ip));
+  // TODO(STM): fix this merge
+  // ASSERT(!sh4_r8.is(sh4_rtmp) && !sh4_r9.is(sh4_rtmp) &&
+  //        !sh4_r10.is(sh4_rtmp));
 
   if (do_gc) {
     // Passing r0.
     __ PrepareCallCFunction(1, 0, r1);
+    __ mov(r4, r0);
     __ CallCFunction(ExternalReference::perform_gc_function(isolate),
         1, 0);
   }
 
   ExternalReference scope_depth =
       ExternalReference::heap_always_allocate_scope_depth(isolate);
   if (always_allocate) {
     __ mov(r0, Operand(scope_depth));
     __ ldr(r1, MemOperand(r0));
     __ add(r1, r1, Operand(1));
     __ str(r1, MemOperand(r0));
   }
 
   // Call C built-in.
-  // r0 = argc, r1 = argv
-  __ mov(r0, Operand(r4));
-  __ mov(r1, Operand(r6));
+  // r4 = argc, r5 = argv, r6 = isolate
+  // __ mov(r4, sh4_r8);
+  // __ mov(r5, sh4_r10);
+  // sh4: ref to ::Generate that stored into the stack
+  __ ldr(r4, MemOperand(sp, (1+0)*kPointerSize));
+  __ ldr(r5, MemOperand(sp, (1+2)*kPointerSize));
 
-#if defined(V8_HOST_ARCH_ARM)
-  int frame_alignment = MacroAssembler::ActivationFrameAlignment();
+#if defined(V8_HOST_ARCH_SH4)
+  int frame_alignment = OS::ActivationFrameAlignment();
   int frame_alignment_mask = frame_alignment - 1;
   if (FLAG_debug_code) {
     if (frame_alignment > kPointerSize) {
       Label alignment_as_expected;
       ASSERT(IsPowerOf2(frame_alignment));
       __ tst(sp, Operand(frame_alignment_mask));
       __ b(eq, &alignment_as_expected);
       // Don't use Check here, as it will call Runtime_Abort re-entering here.
       __ stop("Unexpected alignment");
       __ bind(&alignment_as_expected);
     }
   }
 #endif
 
-  __ mov(r2, Operand(ExternalReference::isolate_address()));
+  __ mov(r6, Operand(ExternalReference::isolate_address()));
+
+  // sh4: ref to ::Generate() that stored the builtin into the stack
+  __ ldr(r2, MemOperand(sp, (1+1)*kPointerSize));
 
   // To let the GC traverse the return address of the exit frames, we need to
   // know where the return address is. The CEntryStub is unmovable, so
   // we can store the address on the stack to be able to find it again and
   // we never have to restore it, because it will not change.
   // Compute the return address in lr to return to after the jump below. Pc is
   // already at '+ 8' from the current instruction but return is after three
   // instructions so add another 4 to pc to get the return address.
-  {
-    // Prevent literal pool emission before return address.
-    Assembler::BlockConstPoolScope block_const_pool(masm);
-    masm->add(lr, pc, Operand(4));
-    __ str(lr, MemOperand(sp, 0));
-    masm->Jump(r5);
-  }
+
+  // Compute the return address in pr to return to after the jsr below.
+  // We use the addpc operation for this with an offset of 6.
+  // We add 3 * kInstrSize to the pc after the addpc for the size of
+  // the sequence: [str, jsr, nop(delay slot)].
+  __ addpc(r3, 3 * Assembler::kInstrSize, pr);
+#ifdef DEBUG
+  int old_pc = masm->pc_offset();
+#endif
+  __ str(r3, MemOperand(sp, 0));
+  __ jsr(r2);
+  //  __ jsr(sh4_r9);
+#ifdef DEBUG
+  ASSERT(masm->pc_offset() - old_pc == 3 * Assembler::kInstrSize);
+#endif
 
   if (always_allocate) {
     // It's okay to clobber r2 and r3 here. Don't mess with r0 and r1
     // though (contain the result).
     __ mov(r2, Operand(scope_depth));
     __ ldr(r3, MemOperand(r2));
     __ sub(r3, r3, Operand(1));
     __ str(r3, MemOperand(r2));
   }
 
   // check for failure result
   Label failure_returned;
   STATIC_ASSERT(((kFailureTag + 1) & kFailureTagMask) == 0);
   // Lower 2 bits of r2 are 0 iff r0 has failure tag.
   __ add(r2, r0, Operand(1));
   __ tst(r2, Operand(kFailureTagMask));
   __ b(eq, &failure_returned);
 
   // Exit C frame and return.
   // r0:r1: result
   // sp: stack pointer
   // fp: frame pointer
-  //  Callee-saved register r4 still holds argc.
-  __ LeaveExitFrame(save_doubles_, r4);
-  __ mov(pc, lr);
+  //  Callee-saved register sh4_r8 still holds argc.
+  // sh4: stored on stack into ::Generate()
+  __ ldr(r2, MemOperand(sp, (1+0)*kPointerSize));
+  __ LeaveExitFrame(save_doubles_, r2);
+  //  __ LeaveExitFrame(save_doubles_, sh4_r8);
+  __ rts();
 
   // check if we should retry or throw exception
   Label retry;
   __ bind(&failure_returned);
   STATIC_ASSERT(Failure::RETRY_AFTER_GC == 0);
   __ tst(r0, Operand(((1 << kFailureTypeTagSize) - 1) << kFailureTagSize));
-  __ b(eq, &retry);
+  __ b(eq, &retry, Label::kNear);
 
   // Special handling of out of memory exceptions.
   Failure* out_of_memory = Failure::OutOfMemoryException();
   __ cmp(r0, Operand(reinterpret_cast<int32_t>(out_of_memory)));
   __ b(eq, throw_out_of_memory_exception);
 
   // Retrieve the pending exception and clear the variable.
   __ mov(r3, Operand(isolate->factory()->the_hole_value()));
   __ mov(ip, Operand(ExternalReference(Isolate::kPendingExceptionAddress,
                                        isolate)));
   __ ldr(r0, MemOperand(ip));
   __ str(r3, MemOperand(ip));
 
   // Special handling of termination exceptions which are uncatchable
   // by javascript code.
-  __ cmp(r0, Operand(isolate->factory()->termination_exception()));
+  __ mov(r3, Operand(isolate->factory()->termination_exception()));
+  __ cmpeq(r0, r3);
   __ b(eq, throw_termination_exception);
 
   // Handle normal exception.
   __ jmp(throw_normal_exception);
 
   __ bind(&retry);  // pass last failure (r0) as parameter (r0) when retrying
 }
 
 
 void CEntryStub::Generate(MacroAssembler* masm) {
   // Called from JavaScript; parameters are on stack as if calling JS function
   // r0: number of arguments including receiver
   // r1: pointer to builtin function
   // fp: frame pointer  (restored after C call)
   // sp: stack pointer  (restored as callee's sp after C call)
   // cp: current context  (C callee-saved)
 
+  // sh4: clobbers r3
   // Result returned in r0 or r0+r1 by default.
 
   // NOTE: Invocations of builtins may return failure objects
   // instead of a proper result. The builtin entry handles
   // this by performing a garbage collection and retrying the
   // builtin once.
 
   // Compute the argv pointer in a callee-saved register.
-  __ add(r6, sp, Operand(r0, LSL, kPointerSizeLog2));
-  __ sub(r6, r6, Operand(kPointerSize));
+  // sh4: will be saved on stack
+  // __ lsl(sh4_r10, r0, Operand(kPointerSizeLog2));
+  // __ add(sh4_r10, sp, sh4_r10);
+  // __ sub(sh4_r10, sh4_r10, Operand(kPointerSize));
+  __ lsl(r3, r0, Operand(kPointerSizeLog2));
+  __ add(r3, sp, r3);
+  __ sub(r3, r3, Operand(kPointerSize));
 
   // Enter the exit frame that transitions from JavaScript to C++.
   FrameScope scope(masm, StackFrame::MANUAL);
-  __ EnterExitFrame(save_doubles_);
+  // SH4: Reserve space for 3 stack locations
+  __ EnterExitFrame(save_doubles_, 3);
 
-  // Set up argc and the builtin function in callee-saved registers.
-  __ mov(r4, Operand(r0));
-  __ mov(r5, Operand(r1));
+  // Setup argc and the builtin function in callee-saved registers.  sh4: save
+  // on stack instead of keep in callee-saved sh4: sp contains: sp[0] == lr;
+  // sp[1] == argc; sp[2] == builtin; sp[3] = argv
+  // __ mov(sh4_r8, r0);
+  // __ mov(sh4_r9, r1);
+  __ str(r0, MemOperand(sp, (1+0)*kPointerSize));  // skip lr location at sp[1]
+  __ str(r1, MemOperand(sp, (1+1)*kPointerSize));
+  __ str(r3, MemOperand(sp, (1+2)*kPointerSize));
 
-  // r4: number of arguments (C callee-saved)
-  // r5: pointer to builtin function (C callee-saved)
-  // r6: pointer to first argument (C callee-saved)
+  // sh4_r8: number of arguments (C callee-saved)
+  // sh4_r9: pointer to builtin function (C callee-saved)
+  // sh4_r10: pointer to first argument (C callee-saved)
+  // ASSERT(!sh4_r8.is(sh4_rtmp) && !sh4_r9.is(sh4_rtmp) &&
+  //        !sh4_r10.is(sh4_rtmp));
 
   Label throw_normal_exception;
   Label throw_termination_exception;
   Label throw_out_of_memory_exception;
 
   // Call into the runtime system.
   GenerateCore(masm,
                &throw_normal_exception,
                &throw_termination_exception,
                &throw_out_of_memory_exception,
@@ skipping 37 matching lines @@
 
   __ bind(&throw_termination_exception);
   __ ThrowUncatchable(r0);
 
   __ bind(&throw_normal_exception);
   __ Throw(r0);
 }
 
 
 void JSEntryStub::GenerateBody(MacroAssembler* masm, bool is_construct) {
-  // r0: code entry
-  // r1: function
-  // r2: receiver
-  // r3: argc
+  // r4: code entry
+  // r5: function
+  // r6: receiver
+  // r7: argc
   // [sp+0]: argv
 
   Label invoke, handler_entry, exit;
 
-  // Called from C, so do not pop argc and args on exit (preserve sp)
-  // No need to save register-passed args
-  // Save callee-saved registers (incl. cp and fp), sp, and lr
-  __ stm(db_w, sp, kCalleeSaved | lr.bit());
+  // Save callee-saved registers
+  __ push(pr);
+  __ pushm(kCalleeSaved);
 
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    // Save callee-saved vfp registers.
-    __ vstm(db_w, sp, kFirstCalleeSavedDoubleReg, kLastCalleeSavedDoubleReg);
-    // Set up the reserved register for 0.0.
-    __ vmov(kDoubleRegZero, 0.0);
-  }
+  // We don't need to save the callee saved double registers: we only use the
+  // caller saved ones.
 
-  // Get address of argv, see stm above.
+  // Move the registers to use ARM ABI (and JS ABI)
+  __ mov(r0, r4);
+  __ mov(r1, r5);
+  __ mov(r2, r6);
+  __ mov(r3, r7);
+
+  // Get address of argv
   // r0: code entry
   // r1: function
   // r2: receiver
   // r3: argc
 
-  // Set up argv in r4.
+  // Setup argv in r4.
   int offset_to_argv = (kNumCalleeSaved + 1) * kPointerSize;
-  if (CpuFeatures::IsSupported(VFP2)) {
-    offset_to_argv += kNumDoubleCalleeSaved * kDoubleSize;
-  }
   __ ldr(r4, MemOperand(sp, offset_to_argv));
 
   // Push a frame with special values setup to mark it as an entry frame.
   // r0: code entry
   // r1: function
   // r2: receiver
   // r3: argc
   // r4: argv
   Isolate* isolate = masm->isolate();
-  __ mov(r8, Operand(-1));  // Push a bad frame pointer to fail if it is used.
+  // Push a bad frame pointer to fail if it is used.
+  __ mov(ip, Operand(-1));
+  __ push(ip);
+
   int marker = is_construct ? StackFrame::ENTRY_CONSTRUCT : StackFrame::ENTRY;
   __ mov(r7, Operand(Smi::FromInt(marker)));
   __ mov(r6, Operand(Smi::FromInt(marker)));
   __ mov(r5,
          Operand(ExternalReference(Isolate::kCEntryFPAddress, isolate)));
   __ ldr(r5, MemOperand(r5));
-  __ Push(r8, r7, r6, r5);
+  __ push(r7);
+  __ push(r6);
+  __ push(r5);
 
-  // Set up frame pointer for the frame to be pushed.
+  // Setup frame pointer for the frame to be pushed.
   __ add(fp, sp, Operand(-EntryFrameConstants::kCallerFPOffset));
 
   // If this is the outermost JS call, set js_entry_sp value.
   Label non_outermost_js;
   ExternalReference js_entry_sp(Isolate::kJSEntrySPAddress, isolate);
   __ mov(r5, Operand(ExternalReference(js_entry_sp)));
   __ ldr(r6, MemOperand(r5));
-  __ cmp(r6, Operand::Zero());
-  __ b(ne, &non_outermost_js);
+  __ cmp(r6, Operand(0));
+  __ b(ne, &non_outermost_js, Label::kNear);
   __ str(fp, MemOperand(r5));
   __ mov(ip, Operand(Smi::FromInt(StackFrame::OUTERMOST_JSENTRY_FRAME)));
   Label cont;
-  __ b(&cont);
+  __ b_near(&cont);
   __ bind(&non_outermost_js);
   __ mov(ip, Operand(Smi::FromInt(StackFrame::INNER_JSENTRY_FRAME)));
   __ bind(&cont);
   __ push(ip);
 
   // Jump to a faked try block that does the invoke, with a faked catch
   // block that sets the pending exception.
   __ jmp(&invoke);
 
   // Block literal pool emission whilst taking the position of the handler
   // entry. This avoids making the assumption that literal pools are always
   // emitted after an instruction is emitted, rather than before.
   {
-    Assembler::BlockConstPoolScope block_const_pool(masm);
+    // TODO(STM): block constant pool
     __ bind(&handler_entry);
     handler_offset_ = handler_entry.pos();
     // Caught exception: Store result (exception) in the pending exception
     // field in the JSEnv and return a failure sentinel.  Coming in here the
     // fp will be invalid because the PushTryHandler below sets it to 0 to
     // signal the existence of the JSEntry frame.
     __ mov(ip, Operand(ExternalReference(Isolate::kPendingExceptionAddress,
                                          isolate)));
   }
   __ str(r0, MemOperand(ip));
   __ mov(r0, Operand(reinterpret_cast<int32_t>(Failure::Exception())));
   __ b(&exit);
 
   // Invoke: Link this frame into the handler chain.  There's only one
   // handler block in this code object, so its index is 0.
   __ bind(&invoke);
   // Must preserve r0-r4, r5-r7 are available.
   __ PushTryHandler(StackHandler::JS_ENTRY, 0);
   // If an exception not caught by another handler occurs, this handler
-  // returns control to the code after the bl(&invoke) above, which
+  // returns control to the code after the jmp(&invoke) above, which
   // restores all kCalleeSaved registers (including cp and fp) to their
   // saved values before returning a failure to C.
 
   // Clear any pending exceptions.
   __ mov(r5, Operand(isolate->factory()->the_hole_value()));
   __ mov(ip, Operand(ExternalReference(Isolate::kPendingExceptionAddress,
                                        isolate)));
   __ str(r5, MemOperand(ip));
 
   // Invoke the function by calling through JS entry trampoline builtin.
   // Notice that we cannot store a reference to the trampoline code directly in
   // this stub, because runtime stubs are not traversed when doing GC.
 
   // Expected registers by Builtins::JSEntryTrampoline
   // r0: code entry
   // r1: function
   // r2: receiver
   // r3: argc
   // r4: argv
   if (is_construct) {
     ExternalReference construct_entry(Builtins::kJSConstructEntryTrampoline,
                                       isolate);
     __ mov(ip, Operand(construct_entry));
   } else {
     ExternalReference entry(Builtins::kJSEntryTrampoline, isolate);
     __ mov(ip, Operand(entry));
   }
   __ ldr(ip, MemOperand(ip));  // deref address
 
-  // Branch and link to JSEntryTrampoline.  We don't use the double underscore
-  // macro for the add instruction because we don't want the coverage tool
-  // inserting instructions here after we read the pc. We block literal pool
-  // emission for the same reason.
-  {
-    Assembler::BlockConstPoolScope block_const_pool(masm);
-    __ mov(lr, Operand(pc));
-    masm->add(pc, ip, Operand(Code::kHeaderSize - kHeapObjectTag));
-  }
+  // JSEntryTrampoline
+  __ add(ip, Operand(Code::kHeaderSize - kHeapObjectTag));
+  __ jsr(ip);
 
   // Unlink this frame from the handler chain.
   __ PopTryHandler();
 
   __ bind(&exit);  // r0 holds result
   // Check if the current stack frame is marked as the outermost JS frame.
   Label non_outermost_js_2;
   __ pop(r5);
   __ cmp(r5, Operand(Smi::FromInt(StackFrame::OUTERMOST_JSENTRY_FRAME)));
-  __ b(ne, &non_outermost_js_2);
-  __ mov(r6, Operand::Zero());
+  __ b(ne, &non_outermost_js_2, Label::kNear);
+  __ mov(r6, Operand(0));
   __ mov(r5, Operand(ExternalReference(js_entry_sp)));
   __ str(r6, MemOperand(r5));
   __ bind(&non_outermost_js_2);
 
   // Restore the top frame descriptors from the stack.
   __ pop(r3);
   __ mov(ip,
          Operand(ExternalReference(Isolate::kCEntryFPAddress, isolate)));
   __ str(r3, MemOperand(ip));
 
   // Reset the stack to the callee saved registers.
   __ add(sp, sp, Operand(-EntryFrameConstants::kCallerFPOffset));
 
   // Restore callee-saved registers and return.
-#ifdef DEBUG
-  if (FLAG_debug_code) {
-    __ mov(lr, Operand(pc));
-  }
-#endif
+  __ popm(kCalleeSaved);
+  __ pop(pr);
 
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
-    // Restore callee-saved vfp registers.
-    __ vldm(ia_w, sp, kFirstCalleeSavedDoubleReg, kLastCalleeSavedDoubleReg);
-  }
-
-  __ ldm(ia_w, sp, kCalleeSaved | pc.bit());
+  __ rts();
 }
 
 
 // Uses registers r0 to r4.
 // Expected input (depending on whether args are in registers or on the stack):
 // * object: r0 or at sp + 1 * kPointerSize.
 // * function: r1 or at sp.
 //
 // An inlined call site may have been generated before calling this stub.
 // In this case the offset to the inline site to patch is passed on the stack,
@@ skipping 70 matching lines @@
   // Get prototype of object into r2.
   __ ldr(scratch, FieldMemOperand(map, Map::kPrototypeOffset));
 
   // We don't need map any more. Use it as a scratch register.
   Register scratch2 = map;
   map = no_reg;
 
   // Loop through the prototype chain looking for the function prototype.
   __ LoadRoot(scratch2, Heap::kNullValueRootIndex);
   __ bind(&loop);
-  __ cmp(scratch, Operand(prototype));
-  __ b(eq, &is_instance);
+  __ cmp(scratch, prototype);
+  __ b(eq, &is_instance, Label::kNear);
   __ cmp(scratch, scratch2);
   __ b(eq, &is_not_instance);
   __ ldr(scratch, FieldMemOperand(scratch, HeapObject::kMapOffset));
   __ ldr(scratch, FieldMemOperand(scratch, Map::kPrototypeOffset));
   __ jmp(&loop);
 
   __ bind(&is_instance);
   if (!HasCallSiteInlineCheck()) {
     __ mov(r0, Operand(Smi::FromInt(0)));
     __ StoreRoot(r0, Heap::kInstanceofCacheAnswerRootIndex);
@@ skipping 27 matching lines @@
       __ mov(r0, Operand(Smi::FromInt(1)));
     }
   }
   __ Ret(HasArgsInRegisters() ? 0 : 2);
 
   Label object_not_null, object_not_null_or_smi;
   __ bind(&not_js_object);
   // Before null, smi and string value checks, check that the rhs is a function
   // as for a non-function rhs an exception needs to be thrown.
   __ JumpIfSmi(function, &slow);
-  __ CompareObjectType(function, scratch2, scratch, JS_FUNCTION_TYPE);
+  __ CompareObjectType(function, scratch2, scratch, JS_FUNCTION_TYPE, eq);
   __ b(ne, &slow);
 
   // Null is not instance of anything.
   __ cmp(scratch, Operand(masm->isolate()->factory()->null_value()));
-  __ b(ne, &object_not_null);
+  __ b(ne, &object_not_null, Label::kNear);
   __ mov(r0, Operand(Smi::FromInt(1)));
   __ Ret(HasArgsInRegisters() ? 0 : 2);
 
   __ bind(&object_not_null);
   // Smi values are not instances of anything.
   __ JumpIfNotSmi(object, &object_not_null_or_smi);
   __ mov(r0, Operand(Smi::FromInt(1)));
   __ Ret(HasArgsInRegisters() ? 0 : 2);
 
   __ bind(&object_not_null_or_smi);
   // String values are not instances of anything.
   __ IsObjectJSStringType(object, scratch, &slow);
   __ mov(r0, Operand(Smi::FromInt(1)));
   __ Ret(HasArgsInRegisters() ? 0 : 2);
 
   // Slow-case.  Tail call builtin.
   __ bind(&slow);
   if (!ReturnTrueFalseObject()) {
     if (HasArgsInRegisters()) {
       __ Push(r0, r1);
     }
   __ InvokeBuiltin(Builtins::INSTANCE_OF, JUMP_FUNCTION);
   } else {
     {
       FrameScope scope(masm, StackFrame::INTERNAL);
       __ Push(r0, r1);
       __ InvokeBuiltin(Builtins::INSTANCE_OF, CALL_FUNCTION);
     }
-    __ cmp(r0, Operand::Zero());
-    __ LoadRoot(r0, Heap::kTrueValueRootIndex, eq);
-    __ LoadRoot(r0, Heap::kFalseValueRootIndex, ne);
+    Label ltrue, lfalse;
+    __ cmp(r0, Operand(0));
+    __ bf_near(&lfalse);
+    __ LoadRoot(r0, Heap::kTrueValueRootIndex);
+    __ jmp_near(&ltrue);
+    __ bind(&lfalse);
+    __ LoadRoot(r0, Heap::kFalseValueRootIndex);
+    __ bind(&ltrue);
     __ Ret(HasArgsInRegisters() ? 0 : 2);
   }
 }
 
 
 Register InstanceofStub::left() { return r0; }
 
 
 Register InstanceofStub::right() { return r1; }
 
 
 void ArgumentsAccessStub::GenerateReadElement(MacroAssembler* masm) {
   // The displacement is the offset of the last parameter (if any)
   // relative to the frame pointer.
   const int kDisplacement =
       StandardFrameConstants::kCallerSPOffset - kPointerSize;
 
   // Check that the key is a smi.
   Label slow;
-  __ JumpIfNotSmi(r1, &slow);
+  __ JumpIfNotSmi(r1, &slow, Label::kNear);
 
   // Check if the calling frame is an arguments adaptor frame.
   Label adaptor;
   __ ldr(r2, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
   __ ldr(r3, MemOperand(r2, StandardFrameConstants::kContextOffset));
   __ cmp(r3, Operand(Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR)));
-  __ b(eq, &adaptor);
+  __ b(eq, &adaptor, Label::kNear);
 
   // Check index against formal parameters count limit passed in
   // through register r0. Use unsigned comparison to get negative
   // check for free.
-  __ cmp(r1, r0);
-  __ b(hs, &slow);
+  __ cmphs(r1, r0);
+  __ bt_near(&slow);
 
   // Read the argument from the stack and return it.
   __ sub(r3, r0, r1);
-  __ add(r3, fp, Operand(r3, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r0, r3, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r3, fp, r0);
   __ ldr(r0, MemOperand(r3, kDisplacement));
-  __ Jump(lr);
+  __ rts();
 
   // Arguments adaptor case: Check index against actual arguments
   // limit found in the arguments adaptor frame. Use unsigned
   // comparison to get negative check for free.
   __ bind(&adaptor);
   __ ldr(r0, MemOperand(r2, ArgumentsAdaptorFrameConstants::kLengthOffset));
-  __ cmp(r1, r0);
-  __ b(cs, &slow);
+  __ cmphs(r1, r0);
+  __ bt_near(&slow);
 
   // Read the argument from the adaptor frame and return it.
   __ sub(r3, r0, r1);
-  __ add(r3, r2, Operand(r3, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r0, r3, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r3, r2, r0);
   __ ldr(r0, MemOperand(r3, kDisplacement));
-  __ Jump(lr);
+  __ rts();
 
   // Slow-case: Handle non-smi or out-of-bounds access to arguments
   // by calling the runtime system.
   __ bind(&slow);
   __ push(r1);
   __ TailCallRuntime(Runtime::kGetArgumentsProperty, 1, 1);
 }
 
 
 void ArgumentsAccessStub::GenerateNewNonStrictSlow(MacroAssembler* masm) {
   // sp[0] : number of parameters
   // sp[4] : receiver displacement
   // sp[8] : function
 
   // Check if the calling frame is an arguments adaptor frame.
   Label runtime;
   __ ldr(r3, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
   __ ldr(r2, MemOperand(r3, StandardFrameConstants::kContextOffset));
   __ cmp(r2, Operand(Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR)));
-  __ b(ne, &runtime);
+  __ b(ne, &runtime, Label::kNear);
 
   // Patch the arguments.length and the parameters pointer in the current frame.
   __ ldr(r2, MemOperand(r3, ArgumentsAdaptorFrameConstants::kLengthOffset));
   __ str(r2, MemOperand(sp, 0 * kPointerSize));
-  __ add(r3, r3, Operand(r2, LSL, 1));
+  __ lsl(ip, r2, Operand(1));
+  __ add(r3, r3, ip);
   __ add(r3, r3, Operand(StandardFrameConstants::kCallerSPOffset));
   __ str(r3, MemOperand(sp, 1 * kPointerSize));
 
   __ bind(&runtime);
   __ TailCallRuntime(Runtime::kNewArgumentsFast, 3, 1);
 }
 
 
 void ArgumentsAccessStub::GenerateNewNonStrictFast(MacroAssembler* masm) {
   // Stack layout:
   //  sp[0] : number of parameters (tagged)
   //  sp[4] : address of receiver argument
   //  sp[8] : function
   // Registers used over whole function:
   //  r6 : allocated object (tagged)
   //  r9 : mapped parameter count (tagged)
 
   __ ldr(r1, MemOperand(sp, 0 * kPointerSize));
   // r1 = parameter count (tagged)
 
   // Check if the calling frame is an arguments adaptor frame.
   Label runtime;
   Label adaptor_frame, try_allocate;
   __ ldr(r3, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
   __ ldr(r2, MemOperand(r3, StandardFrameConstants::kContextOffset));
   __ cmp(r2, Operand(Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR)));
-  __ b(eq, &adaptor_frame);
+  __ b(eq, &adaptor_frame, Label::kNear);
 
   // No adaptor, parameter count = argument count.
   __ mov(r2, r1);
-  __ b(&try_allocate);
+  __ b_near(&try_allocate);
 
   // We have an adaptor frame. Patch the parameters pointer.
   __ bind(&adaptor_frame);
   __ ldr(r2, MemOperand(r3, ArgumentsAdaptorFrameConstants::kLengthOffset));
-  __ add(r3, r3, Operand(r2, LSL, 1));
+  __ lsl(ip, r2, Operand(1));
+  __ add(r3, r3, ip);
   __ add(r3, r3, Operand(StandardFrameConstants::kCallerSPOffset));
   __ str(r3, MemOperand(sp, 1 * kPointerSize));
 
   // r1 = parameter count (tagged)
   // r2 = argument count (tagged)
   // Compute the mapped parameter count = min(r1, r2) in r1.
-  __ cmp(r1, Operand(r2));
-  __ mov(r1, Operand(r2), LeaveCC, gt);
+  __ cmpgt(r1, r2);
+  __ mov(r1, r2, t);
 
   __ bind(&try_allocate);
 
   // Compute the sizes of backing store, parameter map, and arguments object.
   // 1. Parameter map, has 2 extra words containing context and backing store.
   const int kParameterMapHeaderSize =
       FixedArray::kHeaderSize + 2 * kPointerSize;
   // If there are no mapped parameters, we do not need the parameter_map.
-  __ cmp(r1, Operand(Smi::FromInt(0)));
-  __ mov(r9, Operand::Zero(), LeaveCC, eq);
-  __ mov(r9, Operand(r1, LSL, 1), LeaveCC, ne);
-  __ add(r9, r9, Operand(kParameterMapHeaderSize), LeaveCC, ne);
+  __ cmpeq(r1, Operand(Smi::FromInt(0)));
+  __ mov(r9, Operand(0), t);
+  Label skip;
+  __ bt_near(&skip);
+  __ lsl(r9, r1, Operand(1));
+  __ add(r9, r9, Operand(kParameterMapHeaderSize));
+  __ bind(&skip);
 
   // 2. Backing store.
-  __ add(r9, r9, Operand(r2, LSL, 1));
+  __ lsl(ip, r2, Operand(1));
+  __ add(r9, r9, ip);
   __ add(r9, r9, Operand(FixedArray::kHeaderSize));
 
   // 3. Arguments object.
   __ add(r9, r9, Operand(Heap::kArgumentsObjectSize));
 
   // Do the allocation of all three objects in one go.
   __ AllocateInNewSpace(r9, r0, r3, r4, &runtime, TAG_OBJECT);
 
   // r0 = address of new object(s) (tagged)
   // r2 = argument count (tagged)
   // Get the arguments boilerplate from the current native context into r4.
   const int kNormalOffset =
       Context::SlotOffset(Context::ARGUMENTS_BOILERPLATE_INDEX);
   const int kAliasedOffset =
       Context::SlotOffset(Context::ALIASED_ARGUMENTS_BOILERPLATE_INDEX);
 
-  __ ldr(r4, MemOperand(r8, Context::SlotOffset(Context::GLOBAL_OBJECT_INDEX)));
+  __ ldr(r4, MemOperand(cp, Context::SlotOffset(Context::GLOBAL_OBJECT_INDEX)));
   __ ldr(r4, FieldMemOperand(r4, GlobalObject::kNativeContextOffset));
-  __ cmp(r1, Operand::Zero());
-  __ ldr(r4, MemOperand(r4, kNormalOffset), eq);
-  __ ldr(r4, MemOperand(r4, kAliasedOffset), ne);
+  __ cmp(r1, Operand(0));
+  Label lf, end;
+  __ bf_near(&lf);
+  __ ldr(r4, MemOperand(r4, kNormalOffset));
+  __ b_near(&end);
+  __ bind(&lf);
+  __ ldr(r4, MemOperand(r4, kAliasedOffset));
+  __ bind(&end);
 
   // r0 = address of new object (tagged)
   // r1 = mapped parameter count (tagged)
   // r2 = argument count (tagged)
   // r4 = address of boilerplate object (tagged)
   // Copy the JS object part.
   for (int i = 0; i < JSObject::kHeaderSize; i += kPointerSize) {
     __ ldr(r3, FieldMemOperand(r4, i));
     __ str(r3, FieldMemOperand(r0, i));
   }
@@ skipping 19 matching lines @@
 
   // r0 = address of new object (tagged)
   // r1 = mapped parameter count (tagged)
   // r2 = argument count (tagged)
   // r4 = address of parameter map or backing store (tagged)
   // Initialize parameter map. If there are no mapped arguments, we're done.
   Label skip_parameter_map;
   __ cmp(r1, Operand(Smi::FromInt(0)));
   // Move backing store address to r3, because it is
   // expected there when filling in the unmapped arguments.
-  __ mov(r3, r4, LeaveCC, eq);
+  __ mov(r3, r4, eq);
   __ b(eq, &skip_parameter_map);
 
   __ LoadRoot(r6, Heap::kNonStrictArgumentsElementsMapRootIndex);
   __ str(r6, FieldMemOperand(r4, FixedArray::kMapOffset));
   __ add(r6, r1, Operand(Smi::FromInt(2)));
   __ str(r6, FieldMemOperand(r4, FixedArray::kLengthOffset));
-  __ str(r8, FieldMemOperand(r4, FixedArray::kHeaderSize + 0 * kPointerSize));
-  __ add(r6, r4, Operand(r1, LSL, 1));
+  __ str(cp, FieldMemOperand(r4, FixedArray::kHeaderSize + 0 * kPointerSize));
+  __ lsl(r6, r1, Operand(1));
+  __ add(r6, r4, r6);
   __ add(r6, r6, Operand(kParameterMapHeaderSize));
   __ str(r6, FieldMemOperand(r4, FixedArray::kHeaderSize + 1 * kPointerSize));
 
   // Copy the parameter slots and the holes in the arguments.
   // We need to fill in mapped_parameter_count slots. They index the context,
   // where parameters are stored in reverse order, at
   //   MIN_CONTEXT_SLOTS .. MIN_CONTEXT_SLOTS+parameter_count-1
   // The mapped parameter thus need to get indices
   //   MIN_CONTEXT_SLOTS+parameter_count-1 ..
   //       MIN_CONTEXT_SLOTS+parameter_count-mapped_parameter_count
   // We loop from right to left.
   Label parameters_loop, parameters_test;
   __ mov(r6, r1);
   __ ldr(r9, MemOperand(sp, 0 * kPointerSize));
   __ add(r9, r9, Operand(Smi::FromInt(Context::MIN_CONTEXT_SLOTS)));
-  __ sub(r9, r9, Operand(r1));
+  __ sub(r9, r9, r1);
   __ LoadRoot(r7, Heap::kTheHoleValueRootIndex);
-  __ add(r3, r4, Operand(r6, LSL, 1));
+  __ lsl(r3, r6, Operand(1));
+  __ add(r3, r4, r3);
   __ add(r3, r3, Operand(kParameterMapHeaderSize));
 
   // r6 = loop variable (tagged)
   // r1 = mapping index (tagged)
   // r3 = address of backing store (tagged)
   // r4 = address of parameter map (tagged)
   // r5 = temporary scratch (a.o., for address calculation)
   // r7 = the hole value
-  __ jmp(&parameters_test);
+  __ jmp_near(&parameters_test);
 
   __ bind(&parameters_loop);
   __ sub(r6, r6, Operand(Smi::FromInt(1)));
-  __ mov(r5, Operand(r6, LSL, 1));
+  __ lsl(r5, r6, Operand(1));
   __ add(r5, r5, Operand(kParameterMapHeaderSize - kHeapObjectTag));
   __ str(r9, MemOperand(r4, r5));
   __ sub(r5, r5, Operand(kParameterMapHeaderSize - FixedArray::kHeaderSize));
   __ str(r7, MemOperand(r3, r5));
   __ add(r9, r9, Operand(Smi::FromInt(1)));
   __ bind(&parameters_test);
   __ cmp(r6, Operand(Smi::FromInt(0)));
   __ b(ne, &parameters_loop);
 
   __ bind(&skip_parameter_map);
   // r2 = argument count (tagged)
   // r3 = address of backing store (tagged)
   // r5 = scratch
   // Copy arguments header and remaining slots (if there are any).
   __ LoadRoot(r5, Heap::kFixedArrayMapRootIndex);
   __ str(r5, FieldMemOperand(r3, FixedArray::kMapOffset));
   __ str(r2, FieldMemOperand(r3, FixedArray::kLengthOffset));
 
   Label arguments_loop, arguments_test;
   __ mov(r9, r1);
   __ ldr(r4, MemOperand(sp, 1 * kPointerSize));
-  __ sub(r4, r4, Operand(r9, LSL, 1));
-  __ jmp(&arguments_test);
+  __ lsl(r5, r9, Operand(1));
+  __ sub(r4, r4, r5);
+  __ jmp_near(&arguments_test);
 
   __ bind(&arguments_loop);
   __ sub(r4, r4, Operand(kPointerSize));
   __ ldr(r6, MemOperand(r4, 0));
-  __ add(r5, r3, Operand(r9, LSL, 1));
+  __ lsl(r5, r9, Operand(1));
+  __ add(r5, r3, r5);
   __ str(r6, FieldMemOperand(r5, FixedArray::kHeaderSize));
   __ add(r9, r9, Operand(Smi::FromInt(1)));
 
   __ bind(&arguments_test);
-  __ cmp(r9, Operand(r2));
-  __ b(lt, &arguments_loop);
+  __ cmpge(r9, r2);
+  __ bf(&arguments_loop);
 
   // Return and remove the on-stack parameters.
   __ add(sp, sp, Operand(3 * kPointerSize));
   __ Ret();
 
   // Do the runtime call to allocate the arguments object.
   // r2 = argument count (tagged)
   __ bind(&runtime);
   __ str(r2, MemOperand(sp, 0 * kPointerSize));  // Patch argument count.
   __ TailCallRuntime(Runtime::kNewArgumentsFast, 3, 1);
 }
 
 
 void ArgumentsAccessStub::GenerateNewStrict(MacroAssembler* masm) {
   // sp[0] : number of parameters
   // sp[4] : receiver displacement
   // sp[8] : function
   // Check if the calling frame is an arguments adaptor frame.
   Label adaptor_frame, try_allocate, runtime;
   __ ldr(r2, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
   __ ldr(r3, MemOperand(r2, StandardFrameConstants::kContextOffset));
   __ cmp(r3, Operand(Smi::FromInt(StackFrame::ARGUMENTS_ADAPTOR)));
-  __ b(eq, &adaptor_frame);
+  __ b(eq, &adaptor_frame, Label::kNear);
 
   // Get the length from the frame.
   __ ldr(r1, MemOperand(sp, 0));
-  __ b(&try_allocate);
+  __ b_near(&try_allocate);
 
   // Patch the arguments.length and the parameters pointer.
   __ bind(&adaptor_frame);
   __ ldr(r1, MemOperand(r2, ArgumentsAdaptorFrameConstants::kLengthOffset));
   __ str(r1, MemOperand(sp, 0));
-  __ add(r3, r2, Operand(r1, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r3, r1, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r3, r2, r3);
   __ add(r3, r3, Operand(StandardFrameConstants::kCallerSPOffset));
   __ str(r3, MemOperand(sp, 1 * kPointerSize));
 
   // Try the new space allocation. Start out with computing the size
   // of the arguments object and the elements array in words.
   Label add_arguments_object;
   __ bind(&try_allocate);
   __ cmp(r1, Operand(0, RelocInfo::NONE));
-  __ b(eq, &add_arguments_object);
-  __ mov(r1, Operand(r1, LSR, kSmiTagSize));
+  __ b(eq, &add_arguments_object, Label::kNear);
+  __ lsr(r1, r1, Operand(kSmiTagSize));
   __ add(r1, r1, Operand(FixedArray::kHeaderSize / kPointerSize));
   __ bind(&add_arguments_object);
   __ add(r1, r1, Operand(Heap::kArgumentsObjectSizeStrict / kPointerSize));
 
   // Do the allocation of both objects in one go.
-  __ AllocateInNewSpace(r1,
-                        r0,
-                        r2,
-                        r3,
+  __ AllocateInNewSpace(r1,  // object size
+                        r0,  // result
+                        r2,  // scratch1
+                        r3,  // scratch2
                         &runtime,
                         static_cast<AllocationFlags>(TAG_OBJECT |
                                                      SIZE_IN_WORDS));
 
   // Get the arguments boilerplate from the current native context.
   __ ldr(r4, MemOperand(cp, Context::SlotOffset(Context::GLOBAL_OBJECT_INDEX)));
   __ ldr(r4, FieldMemOperand(r4, GlobalObject::kNativeContextOffset));
   __ ldr(r4, MemOperand(r4, Context::SlotOffset(
       Context::STRICT_MODE_ARGUMENTS_BOILERPLATE_INDEX)));
 
   // Copy the JS object part.
   __ CopyFields(r0, r4, r3.bit(), JSObject::kHeaderSize / kPointerSize);
 
   // Get the length (smi tagged) and set that as an in-object property too.
   STATIC_ASSERT(Heap::kArgumentsLengthIndex == 0);
   __ ldr(r1, MemOperand(sp, 0 * kPointerSize));
   __ str(r1, FieldMemOperand(r0, JSObject::kHeaderSize +
       Heap::kArgumentsLengthIndex * kPointerSize));
 
   // If there are no actual arguments, we're done.
   Label done;
   __ cmp(r1, Operand(0, RelocInfo::NONE));
-  __ b(eq, &done);
+  __ b(eq, &done, Label::kNear);
 
   // Get the parameters pointer from the stack.
   __ ldr(r2, MemOperand(sp, 1 * kPointerSize));
 
-  // Set up the elements pointer in the allocated arguments object and
+  // Setup the elements pointer in the allocated arguments object and
   // initialize the header in the elements fixed array.
   __ add(r4, r0, Operand(Heap::kArgumentsObjectSizeStrict));
   __ str(r4, FieldMemOperand(r0, JSObject::kElementsOffset));
   __ LoadRoot(r3, Heap::kFixedArrayMapRootIndex);
   __ str(r3, FieldMemOperand(r4, FixedArray::kMapOffset));
   __ str(r1, FieldMemOperand(r4, FixedArray::kLengthOffset));
   // Untag the length for the loop.
-  __ mov(r1, Operand(r1, LSR, kSmiTagSize));
+  __ lsr(r1, r1, Operand(kSmiTagSize));
 
   // Copy the fixed array slots.
   Label loop;
-  // Set up r4 to point to the first array slot.
+  // Setup r4 to point to the first array slot.
   __ add(r4, r4, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
   __ bind(&loop);
   // Pre-decrement r2 with kPointerSize on each iteration.
   // Pre-decrement in order to skip receiver.
-  __ ldr(r3, MemOperand(r2, kPointerSize, NegPreIndex));
+  __ sub(r2, r2, Operand(kPointerSize));
+  __ ldr(r3, MemOperand(r2));
   // Post-increment r4 with kPointerSize on each iteration.
   __ str(r3, MemOperand(r4, kPointerSize, PostIndex));
-  __ sub(r1, r1, Operand(1));
-  __ cmp(r1, Operand(0, RelocInfo::NONE));
+  __ dt(r1);
   __ b(ne, &loop);
 
   // Return and remove the on-stack parameters.
   __ bind(&done);
   __ add(sp, sp, Operand(3 * kPointerSize));
   __ Ret();
 
   // Do the runtime call to allocate the arguments object.
   __ bind(&runtime);
   __ TailCallRuntime(Runtime::kNewStrictArgumentsFast, 3, 1);
@@ skipping 38 matching lines @@
       ExternalReference::address_of_regexp_stack_memory_size(isolate);
   __ mov(r0, Operand(address_of_regexp_stack_memory_size));
   __ ldr(r0, MemOperand(r0, 0));
   __ cmp(r0, Operand(0));
   __ b(eq, &runtime);
 
   // Check that the first argument is a JSRegExp object.
   __ ldr(r0, MemOperand(sp, kJSRegExpOffset));
   STATIC_ASSERT(kSmiTag == 0);
   __ JumpIfSmi(r0, &runtime);
-  __ CompareObjectType(r0, r1, r1, JS_REGEXP_TYPE);
+  __ CompareObjectType(r0, r1, r1, JS_REGEXP_TYPE, eq);
   __ b(ne, &runtime);
 
   // Check that the RegExp has been compiled (data contains a fixed array).
   __ ldr(regexp_data, FieldMemOperand(r0, JSRegExp::kDataOffset));
   if (FLAG_debug_code) {
     __ tst(regexp_data, Operand(kSmiTagMask));
     __ Check(ne, "Unexpected type for RegExp data, FixedArray expected");
-    __ CompareObjectType(regexp_data, r0, r0, FIXED_ARRAY_TYPE);
+    __ CompareObjectType(regexp_data, r0, r0, FIXED_ARRAY_TYPE, eq);
     __ Check(eq, "Unexpected type for RegExp data, FixedArray expected");
   }
 
   // regexp_data: RegExp data (FixedArray)
   // Check the type of the RegExp. Only continue if type is JSRegExp::IRREGEXP.
   __ ldr(r0, FieldMemOperand(regexp_data, JSRegExp::kDataTagOffset));
   __ cmp(r0, Operand(Smi::FromInt(JSRegExp::IRREGEXP)));
   __ b(ne, &runtime);
 
   // regexp_data: RegExp data (FixedArray)
   // Check that the number of captures fit in the static offsets vector buffer.
   __ ldr(r2,
          FieldMemOperand(regexp_data, JSRegExp::kIrregexpCaptureCountOffset));
   // Calculate number of capture registers (number_of_captures + 1) * 2. This
   // uses the asumption that smis are 2 * their untagged value.
   STATIC_ASSERT(kSmiTag == 0);
   STATIC_ASSERT(kSmiTagSize + kSmiShiftSize == 1);
   __ add(r2, r2, Operand(2));  // r2 was a smi.
   // Check that the static offsets vector buffer is large enough.
-  __ cmp(r2, Operand(Isolate::kJSRegexpStaticOffsetsVectorSize));
-  __ b(hi, &runtime);
+  __ cmphi(r2, Operand(Isolate::kJSRegexpStaticOffsetsVectorSize));
+  __ bt(&runtime);
 
   // r2: Number of capture registers
   // regexp_data: RegExp data (FixedArray)
   // Check that the second argument is a string.
   __ ldr(subject, MemOperand(sp, kSubjectOffset));
   __ JumpIfSmi(subject, &runtime);
   Condition is_string = masm->IsObjectStringType(subject, r0);
   __ b(NegateCondition(is_string), &runtime);
   // Get the length of the string to r3.
   __ ldr(r3, FieldMemOperand(subject, String::kLengthOffset));
 
   // r2: Number of capture registers
   // r3: Length of subject string as a smi
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
   // Check that the third argument is a positive smi less than the subject
   // string length. A negative value will be greater (unsigned comparison).
   __ ldr(r0, MemOperand(sp, kPreviousIndexOffset));
   __ JumpIfNotSmi(r0, &runtime);
-  __ cmp(r3, Operand(r0));
-  __ b(ls, &runtime);
+  __ cmphi(r3, r0);
+  __ bf(&runtime);
 
   // r2: Number of capture registers
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
   // Check that the fourth object is a JSArray object.
   __ ldr(r0, MemOperand(sp, kLastMatchInfoOffset));
   __ JumpIfSmi(r0, &runtime);
-  __ CompareObjectType(r0, r1, r1, JS_ARRAY_TYPE);
+  __ CompareObjectType(r0, r1, r1, JS_ARRAY_TYPE, eq);
   __ b(ne, &runtime);
   // Check that the JSArray is in fast case.
   __ ldr(last_match_info_elements,
          FieldMemOperand(r0, JSArray::kElementsOffset));
   __ ldr(r0, FieldMemOperand(last_match_info_elements, HeapObject::kMapOffset));
   __ CompareRoot(r0, Heap::kFixedArrayMapRootIndex);
   __ b(ne, &runtime);
   // Check that the last match info has space for the capture registers and the
   // additional information.
   __ ldr(r0,
          FieldMemOperand(last_match_info_elements, FixedArray::kLengthOffset));
   __ add(r2, r2, Operand(RegExpImpl::kLastMatchOverhead));
-  __ cmp(r2, Operand(r0, ASR, kSmiTagSize));
-  __ b(gt, &runtime);
+  __ asr(ip, r0, Operand(kSmiTagSize));
+  __ cmpgt(r2, ip);
+  __ bt(&runtime);
 
   // Reset offset for possibly sliced string.
   __ mov(r9, Operand(0));
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
   // Check the representation and encoding of the subject string.
   Label seq_string;
   __ ldr(r0, FieldMemOperand(subject, HeapObject::kMapOffset));
   __ ldrb(r0, FieldMemOperand(r0, Map::kInstanceTypeOffset));
   // First check for flat string.  None of the following string type tests will
   // succeed if subject is not a string or a short external string.
-  __ and_(r1,
+  __ land(r1,
           r0,
           Operand(kIsNotStringMask |
                   kStringRepresentationMask |
-                  kShortExternalStringMask),
-          SetCC);
+                  kShortExternalStringMask));
+  __ tst(r1, r1);
   STATIC_ASSERT((kStringTag | kSeqStringTag) == 0);
   __ b(eq, &seq_string);
 
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
-  // r1: whether subject is a string and if yes, its string representation
   // Check for flat cons string or sliced string.
   // A flat cons string is a cons string where the second part is the empty
   // string. In that case the subject string is just the first part of the cons
   // string. Also in this case the first part of the cons string is known to be
   // a sequential string or an external string.
   // In the case of a sliced string its offset has to be taken into account.
   Label cons_string, external_string, check_encoding;
   STATIC_ASSERT(kConsStringTag < kExternalStringTag);
   STATIC_ASSERT(kSlicedStringTag > kExternalStringTag);
   STATIC_ASSERT(kIsNotStringMask > kExternalStringTag);
   STATIC_ASSERT(kShortExternalStringTag > kExternalStringTag);
-  __ cmp(r1, Operand(kExternalStringTag));
-  __ b(lt, &cons_string);
-  __ b(eq, &external_string);
+  __ cmpge(r1, Operand(kExternalStringTag));
+  __ bf(&cons_string);
+  __ cmpeq(r1, Operand(kExternalStringTag));
+  __ bt(&external_string);
 
   // Catch non-string subject or short external string.
   STATIC_ASSERT(kNotStringTag != 0 && kShortExternalStringTag !=0);
   __ tst(r1, Operand(kIsNotStringMask | kShortExternalStringMask));
   __ b(ne, &runtime);
 
   // String is sliced.
   __ ldr(r9, FieldMemOperand(subject, SlicedString::kOffsetOffset));
-  __ mov(r9, Operand(r9, ASR, kSmiTagSize));
+  __ asr(r9, r9, Operand(kSmiTagSize));
   __ ldr(subject, FieldMemOperand(subject, SlicedString::kParentOffset));
   // r9: offset of sliced string, smi-tagged.
   __ jmp(&check_encoding);
   // String is a cons string, check whether it is flat.
   __ bind(&cons_string);
   __ ldr(r0, FieldMemOperand(subject, ConsString::kSecondOffset));
   __ CompareRoot(r0, Heap::kEmptyStringRootIndex);
   __ b(ne, &runtime);
   __ ldr(subject, FieldMemOperand(subject, ConsString::kFirstOffset));
   // Is first part of cons or parent of slice a flat string?
   __ bind(&check_encoding);
   __ ldr(r0, FieldMemOperand(subject, HeapObject::kMapOffset));
   __ ldrb(r0, FieldMemOperand(r0, Map::kInstanceTypeOffset));
   STATIC_ASSERT(kSeqStringTag == 0);
   __ tst(r0, Operand(kStringRepresentationMask));
   __ b(ne, &external_string);
 
   __ bind(&seq_string);
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
   // r0: Instance type of subject string
   STATIC_ASSERT(4 == kAsciiStringTag);
   STATIC_ASSERT(kTwoByteStringTag == 0);
   // Find the code object based on the assumptions above.
-  __ and_(r0, r0, Operand(kStringEncodingMask));
-  __ mov(r3, Operand(r0, ASR, 2), SetCC);
-  __ ldr(r7, FieldMemOperand(regexp_data, JSRegExp::kDataAsciiCodeOffset), ne);
-  __ ldr(r7, FieldMemOperand(regexp_data, JSRegExp::kDataUC16CodeOffset), eq);
+  __ land(r0, r0, Operand(kStringEncodingMask));
+  __ asr(r3, r0, Operand(2));
+  __ tst(r3, r3);
+  Label skip_true, skip_end;
+  __ bt_near(&skip_true);
+  __ ldr(r7, FieldMemOperand(regexp_data, JSRegExp::kDataAsciiCodeOffset));
+  __ b_near(&skip_end);
+  __ bind(&skip_true);
+  __ ldr(r7, FieldMemOperand(regexp_data, JSRegExp::kDataUC16CodeOffset));
+  __ bind(&skip_end);
 
   // Check that the irregexp code has been generated for the actual string
   // encoding. If it has, the field contains a code object otherwise it contains
   // a smi (code flushing support).
   __ JumpIfSmi(r7, &runtime);
 
   // r3: encoding of subject string (1 if ASCII, 0 if two_byte);
   // r7: code
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
   // Load used arguments before starting to push arguments for call to native
   // RegExp code to avoid handling changing stack height.
   __ ldr(r1, MemOperand(sp, kPreviousIndexOffset));
-  __ mov(r1, Operand(r1, ASR, kSmiTagSize));
+  __ asr(r1, r1, Operand(kSmiTagSize));
 
   // r1: previous index
   // r3: encoding of subject string (1 if ASCII, 0 if two_byte);
   // r7: code
   // subject: Subject string
   // regexp_data: RegExp data (FixedArray)
   // All checks done. Now push arguments for native regexp code.
   __ IncrementCounter(isolate->counters()->regexp_entry_native(), 1, r0, r2);
 
+  // Save r5-r7 as they are going to be used afterward in the C code
+  // r4 is restored by a load on the right place in the same frame
+  __ Push(r5, r6, r7);
+
   // Isolates: note we add an additional parameter here (isolate pointer).
   const int kRegExpExecuteArguments = 9;
   const int kParameterRegisters = 4;
   __ EnterExitFrame(false, kRegExpExecuteArguments - kParameterRegisters);
 
   // Stack pointer now points to cell where return address is to be written.
   // Arguments are before that on the stack or in registers.
 
   // Argument 9 (sp[20]): Pass current isolate address.
   __ mov(r0, Operand(ExternalReference::isolate_address()));
   __ str(r0, MemOperand(sp, 5 * kPointerSize));
 
   // Argument 8 (sp[16]): Indicate that this is a direct call from JavaScript.
   __ mov(r0, Operand(1));
   __ str(r0, MemOperand(sp, 4 * kPointerSize));
 
   // Argument 7 (sp[12]): Start (high end) of backtracking stack memory area.
   __ mov(r0, Operand(address_of_regexp_stack_memory_address));
   __ ldr(r0, MemOperand(r0, 0));
   __ mov(r2, Operand(address_of_regexp_stack_memory_size));
   __ ldr(r2, MemOperand(r2, 0));
-  __ add(r0, r0, Operand(r2));
+  __ add(r0, r0, r2);
   __ str(r0, MemOperand(sp, 3 * kPointerSize));
 
   // Argument 6: Set the number of capture registers to zero to force global
   // regexps to behave as non-global.  This does not affect non-global regexps.
   __ mov(r0, Operand(0));
   __ str(r0, MemOperand(sp, 2 * kPointerSize));
 
   // Argument 5 (sp[4]): static offsets vector buffer.
   __ mov(r0,
          Operand(ExternalReference::address_of_static_offsets_vector(isolate)));
   __ str(r0, MemOperand(sp, 1 * kPointerSize));
 
   // For arguments 4 and 3 get string length, calculate start of string data and
   // calculate the shift of the index (0 for ASCII and 1 for two byte).
-  __ add(r8, subject, Operand(SeqString::kHeaderSize - kHeapObjectTag));
+  __ add(sh4_r8, subject, Operand(SeqString::kHeaderSize - kHeapObjectTag));
   __ eor(r3, r3, Operand(1));
   // Load the length from the original subject string from the previous stack
   // frame. Therefore we have to use fp, which points exactly to two pointer
   // sizes below the previous sp. (Because creating a new stack frame pushes
-  // the previous fp onto the stack and moves up sp by 2 * kPointerSize.)
-  __ ldr(subject, MemOperand(fp, kSubjectOffset + 2 * kPointerSize));
+  // the previous fp onto the stack and moves up sp by 2 * kPointerSize.) We
+  // also have to take into account the 3 registers pushed on the stack
+  // [r5, r7]
+  __ ldr(subject, MemOperand(fp, kSubjectOffset + 2 * kPointerSize +
+                                 3 * kPointerSize));
   // If slice offset is not 0, load the length from the original sliced string.
   // Argument 4, r3: End of string data
   // Argument 3, r2: Start of string data
   // Prepare start and end index of the input.
-  __ add(r9, r8, Operand(r9, LSL, r3));
-  __ add(r2, r9, Operand(r1, LSL, r3));
+  __ lsl(r9, r9, r3);
+  __ add(r9, sh4_r8, r9);
+  __ lsl(ip, r1, r3);
+  __ add(r2, r9, ip);
 
-  __ ldr(r8, FieldMemOperand(subject, String::kLengthOffset));
-  __ mov(r8, Operand(r8, ASR, kSmiTagSize));
-  __ add(r3, r9, Operand(r8, LSL, r3));
+  __ ldr(sh4_r8, FieldMemOperand(subject, String::kLengthOffset));
+  __ asr(sh4_r8, sh4_r8, Operand(kSmiTagSize));
+  __ lsl(r3, sh4_r8, r3);
+  __ add(r3, r9, r3);
 
   // Argument 2 (r1): Previous index.
   // Already there
 
   // Argument 1 (r0): Subject string.
   __ mov(r0, subject);
 
+  __ mov(r4, r0);
+  __ mov(r5, r1);
+  __ mov(r6, r2);
+
   // Locate the code entry and call it.
-  __ add(r7, r7, Operand(Code::kHeaderSize - kHeapObjectTag));
-  DirectCEntryStub stub;
-  stub.GenerateCall(masm, r7);
+  __ add(r0, r7, Operand(Code::kHeaderSize - kHeapObjectTag));
+  __ mov(r7, r3);
 
+  DirectCEntryStub stub(r2);
+  stub.GenerateCall(masm, r0, r3);
+
+  // Get back the subject from the previous frame: r4 will not be scratched by
+  // a call to LeaveExitFrame
+  __ ldr(r4, MemOperand(fp, kSubjectOffset + 2 * kPointerSize +
+                            3 * kPointerSize));
   __ LeaveExitFrame(false, no_reg);
+  __ Pop(r5, r6, r7);
 
   // r0: result
   // subject: subject string (callee saved)
   // regexp_data: RegExp data (callee saved)
   // last_match_info_elements: Last match info elements (callee saved)
 
   // Check the result.
   Label success;
 
   __ cmp(r0, Operand(1));
@@ skipping 41 matching lines @@
   __ ldr(r1,
          FieldMemOperand(regexp_data, JSRegExp::kIrregexpCaptureCountOffset));
   // Calculate number of capture registers (number_of_captures + 1) * 2.
   STATIC_ASSERT(kSmiTag == 0);
   STATIC_ASSERT(kSmiTagSize + kSmiShiftSize == 1);
   __ add(r1, r1, Operand(2));  // r1 was a smi.
 
   // r1: number of capture registers
   // r4: subject string
   // Store the capture count.
-  __ mov(r2, Operand(r1, LSL, kSmiTagSize + kSmiShiftSize));  // To smi.
+  __ lsl(r2, r1, Operand(kSmiTagSize + kSmiShiftSize));  // To smi.
   __ str(r2, FieldMemOperand(last_match_info_elements,
                              RegExpImpl::kLastCaptureCountOffset));
   // Store last subject and last input.
   __ str(subject,
          FieldMemOperand(last_match_info_elements,
                          RegExpImpl::kLastSubjectOffset));
   __ mov(r2, subject);
   __ RecordWriteField(last_match_info_elements,
                       RegExpImpl::kLastSubjectOffset,
                       r2,
@@ skipping 17 matching lines @@
 
   // r1: number of capture registers
   // r2: offsets vector
   Label next_capture, done;
   // Capture register counter starts from number of capture registers and
   // counts down until wraping after zero.
   __ add(r0,
          last_match_info_elements,
          Operand(RegExpImpl::kFirstCaptureOffset - kHeapObjectTag));
   __ bind(&next_capture);
-  __ sub(r1, r1, Operand(1), SetCC);
-  __ b(mi, &done);
+  __ sub(r1, r1, Operand(1));
+  __ cmpge(r1, Operand(0));
+  __ bf(&done);
   // Read the value from the static offsets vector buffer.
   __ ldr(r3, MemOperand(r2, kPointerSize, PostIndex));
   // Store the smi value in the last match info.
-  __ mov(r3, Operand(r3, LSL, kSmiTagSize));
+  __ lsl(r3, r3, Operand(kSmiTagSize));
   __ str(r3, MemOperand(r0, kPointerSize, PostIndex));
   __ jmp(&next_capture);
   __ bind(&done);
 
   // Return last match info.
   __ ldr(r0, MemOperand(sp, kLastMatchInfoOffset));
   __ add(sp, sp, Operand(4 * kPointerSize));
   __ Ret();
 
   // External string.  Short external strings have already been ruled out.
@@ skipping 26 matching lines @@
 void RegExpConstructResultStub::Generate(MacroAssembler* masm) {
   const int kMaxInlineLength = 100;
   Label slowcase;
   Label done;
   Factory* factory = masm->isolate()->factory();
 
   __ ldr(r1, MemOperand(sp, kPointerSize * 2));
   STATIC_ASSERT(kSmiTag == 0);
   STATIC_ASSERT(kSmiTagSize == 1);
   __ JumpIfNotSmi(r1, &slowcase);
-  __ cmp(r1, Operand(Smi::FromInt(kMaxInlineLength)));
-  __ b(hi, &slowcase);
+  __ cmphi(r1, Operand(Smi::FromInt(kMaxInlineLength)));
+  __ b(t, &slowcase);
   // Smi-tagging is equivalent to multiplying by 2.
   // Allocate RegExpResult followed by FixedArray with size in ebx.
   // JSArray:   [Map][empty properties][Elements][Length-smi][index][input]
   // Elements:  [Map][Length][..elements..]
   // Size of JSArray with two in-object properties and the header of a
   // FixedArray.
   int objects_size =
       (JSRegExpResult::kSize + FixedArray::kHeaderSize) / kPointerSize;
-  __ mov(r5, Operand(r1, LSR, kSmiTagSize + kSmiShiftSize));
+  __ lsr(r5, r1, Operand(kSmiTagSize + kSmiShiftSize));
   __ add(r2, r5, Operand(objects_size));
   __ AllocateInNewSpace(
       r2,  // In: Size, in words.
       r0,  // Out: Start of allocation (tagged).
       r3,  // Scratch register.
       r4,  // Scratch register.
       &slowcase,
       static_cast<AllocationFlags>(TAG_OBJECT | SIZE_IN_WORDS));
   // r0: Start of allocated area, object-tagged.
   // r1: Number of elements in array, as smi.
@@ skipping 22 matching lines @@
 
   // Fill out the elements FixedArray.
   // r0: JSArray, tagged.
   // r3: FixedArray, tagged.
   // r5: Number of elements in array, untagged.
 
   // Set map.
   __ mov(r2, Operand(factory->fixed_array_map()));
   __ str(r2, FieldMemOperand(r3, HeapObject::kMapOffset));
   // Set FixedArray length.
-  __ mov(r6, Operand(r5, LSL, kSmiTagSize));
+  __ lsl(r6, r5, Operand(kSmiTagSize));
   __ str(r6, FieldMemOperand(r3, FixedArray::kLengthOffset));
   // Fill contents of fixed-array with undefined.
   __ LoadRoot(r2, Heap::kUndefinedValueRootIndex);
   __ add(r3, r3, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
   // Fill fixed array elements with undefined.
   // r0: JSArray, tagged.
   // r2: undefined.
   // r3: Start of elements in FixedArray.
   // r5: Number of elements to fill.
   Label loop;
-  __ cmp(r5, Operand(0));
+  __ cmpgt(r5, Operand(0));
   __ bind(&loop);
-  __ b(le, &done);  // Jump if r5 is negative or zero.
-  __ sub(r5, r5, Operand(1), SetCC);
-  __ str(r2, MemOperand(r3, r5, LSL, kPointerSizeLog2));
+  __ b(f, &done, Label::kNear);  // Jump if r1 is negative or zero.
+  __ sub(r5, r5, Operand(1));
+  __ lsl(ip, r5, Operand(kPointerSizeLog2));
+  __ str(r2, MemOperand(r3, ip));
+  __ cmpgt(r5, Operand(0));
   __ jmp(&loop);
 
   __ bind(&done);
   __ add(sp, sp, Operand(3 * kPointerSize));
   __ Ret();
 
   __ bind(&slowcase);
   __ TailCallRuntime(Runtime::kRegExpConstructResult, 3, 1);
 }
 
@@ skipping 16 matching lines @@
 
   // A monomorphic cache hit or an already megamorphic state: invoke the
   // function without changing the state.
   __ cmp(r3, r1);
   __ b(eq, &done);
   __ CompareRoot(r3, Heap::kUndefinedValueRootIndex);
   __ b(eq, &done);
 
   // A monomorphic miss (i.e, here the cache is not uninitialized) goes
   // megamorphic.
+  Label skip;
   __ CompareRoot(r3, Heap::kTheHoleValueRootIndex);
   // MegamorphicSentinel is an immortal immovable object (undefined) so no
   // write-barrier is needed.
-  __ LoadRoot(ip, Heap::kUndefinedValueRootIndex, ne);
-  __ str(ip, FieldMemOperand(r2, JSGlobalPropertyCell::kValueOffset), ne);
+  __ bt_near(&skip);
+  __ LoadRoot(ip, Heap::kUndefinedValueRootIndex);
+  __ str(ip, FieldMemOperand(r2, JSGlobalPropertyCell::kValueOffset));
+  __ jmp(&done);
 
   // An uninitialized cache is patched with the function.
-  __ str(r1, FieldMemOperand(r2, JSGlobalPropertyCell::kValueOffset), eq);
+  __ bind(&skip);
+  __ str(r1, FieldMemOperand(r2, JSGlobalPropertyCell::kValueOffset));
   // No need for a write barrier here - cells are rescanned.
 
   __ bind(&done);
 }
 
 
 void CallFunctionStub::Generate(MacroAssembler* masm) {
   // r1 : the function to call
   // r2 : cache cell for call target
   Label slow, non_function;
 
   // The receiver might implicitly be the global object. This is
   // indicated by passing the hole as the receiver to the call
   // function stub.
   if (ReceiverMightBeImplicit()) {
     Label call;
     // Get the receiver from the stack.
     // function, receiver [, arguments]
     __ ldr(r4, MemOperand(sp, argc_ * kPointerSize));
     // Call as function is indicated with the hole.
     __ CompareRoot(r4, Heap::kTheHoleValueRootIndex);
-    __ b(ne, &call);
+    __ b(ne, &call, Label::kNear);
     // Patch the receiver on the stack with the global receiver object.
     __ ldr(r3,
            MemOperand(cp, Context::SlotOffset(Context::GLOBAL_OBJECT_INDEX)));
     __ ldr(r3, FieldMemOperand(r3, GlobalObject::kGlobalReceiverOffset));
     __ str(r3, MemOperand(sp, argc_ * kPointerSize));
     __ bind(&call);
   }
 
   // Check that the function is really a JavaScript function.
   // r1: pushed function (to be verified)
   __ JumpIfSmi(r1, &non_function);
   // Get the map of the function object.
-  __ CompareObjectType(r1, r3, r3, JS_FUNCTION_TYPE);
+  __ CompareObjectType(r1, r2, r2, JS_FUNCTION_TYPE, eq);
   __ b(ne, &slow);
 
   if (RecordCallTarget()) {
     GenerateRecordCallTarget(masm);
   }
 
   // Fast-case: Invoke the function now.
   // r1: pushed function
   ParameterCount actual(argc_);
 
   if (ReceiverMightBeImplicit()) {
     Label call_as_function;
     __ CompareRoot(r4, Heap::kTheHoleValueRootIndex);
-    __ b(eq, &call_as_function);
+    __ b(eq, &call_as_function, Label::kNear);
     __ InvokeFunction(r1,
                       actual,
                       JUMP_FUNCTION,
                       NullCallWrapper(),
                       CALL_AS_METHOD);
     __ bind(&call_as_function);
   }
   __ InvokeFunction(r1,
                     actual,
                     JUMP_FUNCTION,
@@ skipping 40 matching lines @@
 
 void CallConstructStub::Generate(MacroAssembler* masm) {
   // r0 : number of arguments
   // r1 : the function to call
   // r2 : cache cell for call target
   Label slow, non_function_call;
 
   // Check that the function is not a smi.
   __ JumpIfSmi(r1, &non_function_call);
   // Check that the function is a JSFunction.
-  __ CompareObjectType(r1, r3, r3, JS_FUNCTION_TYPE);
+  __ CompareObjectType(r1, r3, r3, JS_FUNCTION_TYPE, eq);
   __ b(ne, &slow);
 
   if (RecordCallTarget()) {
     GenerateRecordCallTarget(masm);
   }
 
   // Jump to the function-specific construct stub.
   __ ldr(r2, FieldMemOperand(r1, JSFunction::kSharedFunctionInfoOffset));
   __ ldr(r2, FieldMemOperand(r2, SharedFunctionInfo::kConstructStubOffset));
-  __ add(pc, r2, Operand(Code::kHeaderSize - kHeapObjectTag));
+  __ add(r2, r2, Operand(Code::kHeaderSize - kHeapObjectTag));
+  __ jmp(r2);
 
   // r0: number of arguments
   // r1: called object
   // r3: object type
   Label do_call;
   __ bind(&slow);
   __ cmp(r3, Operand(JS_FUNCTION_PROXY_TYPE));
   __ b(ne, &non_function_call);
   __ GetBuiltinEntry(r3, Builtins::CALL_FUNCTION_PROXY_AS_CONSTRUCTOR);
   __ jmp(&do_call);
@@ skipping 67 matching lines @@
   // If the receiver is not a string trigger the non-string case.
   __ tst(result_, Operand(kIsNotStringMask));
   __ b(ne, receiver_not_string_);
 
   // If the index is non-smi trigger the non-smi case.
   __ JumpIfNotSmi(index_, &index_not_smi_);
   __ bind(&got_smi_index_);
 
   // Check for index out of range.
   __ ldr(ip, FieldMemOperand(object_, String::kLengthOffset));
-  __ cmp(ip, Operand(index_));
-  __ b(ls, index_out_of_range_);
+  __ cmphi(ip, index_);
+  __ bf(index_out_of_range_);
 
-  __ mov(index_, Operand(index_, ASR, kSmiTagSize));
+  __ asr(index_, index_, Operand(kSmiTagSize));
 
   StringCharLoadGenerator::Generate(masm,
                                     object_,
                                     index_,
                                     result_,
                                     &call_runtime_);
 
-  __ mov(result_, Operand(result_, LSL, kSmiTagSize));
+  __ lsl(result_, result_, Operand(kSmiTagSize));
   __ bind(&exit_);
 }
 
 
 void StringCharCodeAtGenerator::GenerateSlow(
     MacroAssembler* masm,
     const RuntimeCallHelper& call_helper) {
   __ Abort("Unexpected fallthrough to CharCodeAt slow case");
 
   // Index is not a smi.
   __ bind(&index_not_smi_);
   // If index is a heap number, try converting it to an integer.
   __ CheckMap(index_,
               result_,
               Heap::kHeapNumberMapRootIndex,
               index_not_number_,
               DONT_DO_SMI_CHECK);
   call_helper.BeforeCall(masm);
   __ push(object_);
   __ push(index_);  // Consumed by runtime conversion function.
   if (index_flags_ == STRING_INDEX_IS_NUMBER) {
     __ CallRuntime(Runtime::kNumberToIntegerMapMinusZero, 1);
   } else {
     ASSERT(index_flags_ == STRING_INDEX_IS_ARRAY_INDEX);
     // NumberToSmi discards numbers that are not exact integers.
     __ CallRuntime(Runtime::kNumberToSmi, 1);
   }
   // Save the conversion result before the pop instructions below
   // have a chance to overwrite it.
-  __ Move(index_, r0);
+  __ mov(index_, r0);
+
   __ pop(object_);
   // Reload the instance type.
   __ ldr(result_, FieldMemOperand(object_, HeapObject::kMapOffset));
   __ ldrb(result_, FieldMemOperand(result_, Map::kInstanceTypeOffset));
   call_helper.AfterCall(masm);
   // If index is still not a smi, it must be out of range.
   __ JumpIfNotSmi(index_, index_out_of_range_);
   // Otherwise, return to the fast path.
   __ jmp(&got_smi_index_);
 
   // Call runtime. We get here when the receiver is a string and the
   // index is a number, but the code of getting the actual character
   // is too complex (e.g., when the string needs to be flattened).
   __ bind(&call_runtime_);
   call_helper.BeforeCall(masm);
-  __ mov(index_, Operand(index_, LSL, kSmiTagSize));
+  __ lsl(index_, index_, Operand(kSmiTagSize));
   __ Push(object_, index_);
   __ CallRuntime(Runtime::kStringCharCodeAt, 2);
-  __ Move(result_, r0);
+  __ mov(result_, r0);
   call_helper.AfterCall(masm);
   __ jmp(&exit_);
 
   __ Abort("Unexpected fallthrough from CharCodeAt slow case");
 }
 
 
 // -------------------------------------------------------------------------
 // StringCharFromCodeGenerator
 
 void StringCharFromCodeGenerator::GenerateFast(MacroAssembler* masm) {
   // Fast case of Heap::LookupSingleCharacterStringFromCode.
   STATIC_ASSERT(kSmiTag == 0);
   STATIC_ASSERT(kSmiShiftSize == 0);
   ASSERT(IsPowerOf2(String::kMaxAsciiCharCode + 1));
+
+  ASSERT(!code_.is(ip) && !result_.is(ip));
+
   __ tst(code_,
          Operand(kSmiTagMask |
                  ((~String::kMaxAsciiCharCode) << kSmiTagSize)));
   __ b(ne, &slow_case_);
 
   __ LoadRoot(result_, Heap::kSingleCharacterStringCacheRootIndex);
   // At this point code register contains smi tagged ASCII char code.
   STATIC_ASSERT(kSmiTag == 0);
-  __ add(result_, result_, Operand(code_, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(ip, code_, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(result_, result_, ip);
   __ ldr(result_, FieldMemOperand(result_, FixedArray::kHeaderSize));
   __ CompareRoot(result_, Heap::kUndefinedValueRootIndex);
   __ b(eq, &slow_case_);
   __ bind(&exit_);
 }
 
 
 void StringCharFromCodeGenerator::GenerateSlow(
     MacroAssembler* masm,
     const RuntimeCallHelper& call_helper) {
   __ Abort("Unexpected fallthrough to CharFromCode slow case");
 
   __ bind(&slow_case_);
   call_helper.BeforeCall(masm);
   __ push(code_);
   __ CallRuntime(Runtime::kCharFromCode, 1);
-  __ Move(result_, r0);
+  __ mov(result_, r0);
   call_helper.AfterCall(masm);
   __ jmp(&exit_);
 
   __ Abort("Unexpected fallthrough from CharFromCode slow case");
 }
 
 
 // -------------------------------------------------------------------------
 // StringCharAtGenerator
 
@@ skipping 15 matching lines @@
                                           Register dest,
                                           Register src,
                                           Register count,
                                           Register scratch,
                                           bool ascii) {
   Label loop;
   Label done;
   // This loop just copies one character at a time, as it is only used for very
   // short strings.
   if (!ascii) {
-    __ add(count, count, Operand(count), SetCC);
-  } else {
-    __ cmp(count, Operand(0, RelocInfo::NONE));
+    __ add(count, count, count);
   }
-  __ b(eq, &done);
+  __ cmp(count, Operand(0));
+  __ b(eq, &done, Label::kNear);
 
   __ bind(&loop);
-  __ ldrb(scratch, MemOperand(src, 1, PostIndex));
+  __ ldrb(scratch, MemOperand(src));
+  __ add(src, src, Operand(1));
   // Perform sub between load and dependent store to get the load time to
   // complete.
-  __ sub(count, count, Operand(1), SetCC);
-  __ strb(scratch, MemOperand(dest, 1, PostIndex));
+  __ cmpgt(count, Operand(1));
+  __ sub(count, count, Operand(1));
+  __ strb(scratch, MemOperand(dest));
+  __ add(dest, dest, Operand(1));
   // last iteration.
-  __ b(gt, &loop);
+  __ bt(&loop);
 
   __ bind(&done);
 }
 
 
 enum CopyCharactersFlags {
   COPY_ASCII = 1,
   DEST_ALWAYS_ALIGNED = 2
 };
 
@@ skipping 12 matching lines @@
   bool dest_always_aligned = (flags & DEST_ALWAYS_ALIGNED) != 0;
 
   if (dest_always_aligned && FLAG_debug_code) {
     // Check that destination is actually word aligned if the flag says
     // that it is.
     __ tst(dest, Operand(kPointerAlignmentMask));
     __ Check(eq, "Destination of copy not aligned.");
   }
 
   const int kReadAlignment = 4;
-  const int kReadAlignmentMask = kReadAlignment - 1;
   // Ensure that reading an entire aligned word containing the last character
   // of a string will not read outside the allocated area (because we pad up
   // to kObjectAlignment).
   STATIC_ASSERT(kObjectAlignment >= kReadAlignment);
   // Assumes word reads and writes are little endian.
   // Nothing to do for zero characters.
   Label done;
   if (!ascii) {
-    __ add(count, count, Operand(count), SetCC);
-  } else {
-    __ cmp(count, Operand(0, RelocInfo::NONE));
+    __ add(count, count, count);
   }
-  __ b(eq, &done);
+  __ cmpeq(count, Operand(0));
+  __ bt_near(&done);
 
-  // Assume that you cannot read (or write) unaligned.
-  Label byte_loop;
-  // Must copy at least eight bytes, otherwise just do it one byte at a time.
-  __ cmp(count, Operand(8));
-  __ add(count, dest, Operand(count));
-  Register limit = count;  // Read until src equals this.
-  __ b(lt, &byte_loop);
-
-  if (!dest_always_aligned) {
-    // Align dest by byte copying. Copies between zero and three bytes.
-    __ and_(scratch4, dest, Operand(kReadAlignmentMask), SetCC);
-    Label dest_aligned;
-    __ b(eq, &dest_aligned);
-    __ cmp(scratch4, Operand(2));
-    __ ldrb(scratch1, MemOperand(src, 1, PostIndex));
-    __ ldrb(scratch2, MemOperand(src, 1, PostIndex), le);
-    __ ldrb(scratch3, MemOperand(src, 1, PostIndex), lt);
-    __ strb(scratch1, MemOperand(dest, 1, PostIndex));
-    __ strb(scratch2, MemOperand(dest, 1, PostIndex), le);
-    __ strb(scratch3, MemOperand(dest, 1, PostIndex), lt);
-    __ bind(&dest_aligned);
-  }
-
-  Label simple_loop;
-
-  __ sub(scratch4, dest, Operand(src));
-  __ and_(scratch4, scratch4, Operand(0x03), SetCC);
-  __ b(eq, &simple_loop);
-  // Shift register is number of bits in a source word that
-  // must be combined with bits in the next source word in order
-  // to create a destination word.
-
-  // Complex loop for src/dst that are not aligned the same way.
-  {
-    Label loop;
-    __ mov(scratch4, Operand(scratch4, LSL, 3));
-    Register left_shift = scratch4;
-    __ and_(src, src, Operand(~3));  // Round down to load previous word.
-    __ ldr(scratch1, MemOperand(src, 4, PostIndex));
-    // Store the "shift" most significant bits of scratch in the least
-    // signficant bits (i.e., shift down by (32-shift)).
-    __ rsb(scratch2, left_shift, Operand(32));
-    Register right_shift = scratch2;
-    __ mov(scratch1, Operand(scratch1, LSR, right_shift));
-
-    __ bind(&loop);
-    __ ldr(scratch3, MemOperand(src, 4, PostIndex));
-    __ sub(scratch5, limit, Operand(dest));
-    __ orr(scratch1, scratch1, Operand(scratch3, LSL, left_shift));
-    __ str(scratch1, MemOperand(dest, 4, PostIndex));
-    __ mov(scratch1, Operand(scratch3, LSR, right_shift));
-    // Loop if four or more bytes left to copy.
-    // Compare to eight, because we did the subtract before increasing dst.
-    __ sub(scratch5, scratch5, Operand(8), SetCC);
-    __ b(ge, &loop);
-  }
-  // There is now between zero and three bytes left to copy (negative that
-  // number is in scratch5), and between one and three bytes already read into
-  // scratch1 (eight times that number in scratch4). We may have read past
-  // the end of the string, but because objects are aligned, we have not read
-  // past the end of the object.
-  // Find the minimum of remaining characters to move and preloaded characters
-  // and write those as bytes.
-  __ add(scratch5, scratch5, Operand(4), SetCC);
-  __ b(eq, &done);
-  __ cmp(scratch4, Operand(scratch5, LSL, 3), ne);
-  // Move minimum of bytes read and bytes left to copy to scratch4.
-  __ mov(scratch5, Operand(scratch4, LSR, 3), LeaveCC, lt);
-  // Between one and three (value in scratch5) characters already read into
-  // scratch ready to write.
-  __ cmp(scratch5, Operand(2));
-  __ strb(scratch1, MemOperand(dest, 1, PostIndex));
-  __ mov(scratch1, Operand(scratch1, LSR, 8), LeaveCC, ge);
-  __ strb(scratch1, MemOperand(dest, 1, PostIndex), ge);
-  __ mov(scratch1, Operand(scratch1, LSR, 8), LeaveCC, gt);
-  __ strb(scratch1, MemOperand(dest, 1, PostIndex), gt);
-  // Copy any remaining bytes.
-  __ b(&byte_loop);
-
-  // Simple loop.
-  // Copy words from src to dst, until less than four bytes left.
-  // Both src and dest are word aligned.
-  __ bind(&simple_loop);
-  {
-    Label loop;
-    __ bind(&loop);
-    __ ldr(scratch1, MemOperand(src, 4, PostIndex));
-    __ sub(scratch3, limit, Operand(dest));
-    __ str(scratch1, MemOperand(dest, 4, PostIndex));
-    // Compare to 8, not 4, because we do the substraction before increasing
-    // dest.
-    __ cmp(scratch3, Operand(8));
-    __ b(ge, &loop);
-  }
-
-  // Copy bytes from src to dst until dst hits limit.
-  __ bind(&byte_loop);
-  __ cmp(dest, Operand(limit));
-  __ ldrb(scratch1, MemOperand(src, 1, PostIndex), lt);
-  __ b(ge, &done);
-  __ strb(scratch1, MemOperand(dest, 1, PostIndex));
-  __ b(&byte_loop);
-
+  // Use an optimized version for sh4
+  __ memcpy(dest, src, count, scratch1, scratch2, scratch3, scratch4);
   __ bind(&done);
 }
 
 
 void StringHelper::GenerateTwoCharacterSymbolTableProbe(MacroAssembler* masm,
                                                         Register c1,
                                                         Register c2,
                                                         Register scratch1,
                                                         Register scratch2,
                                                         Register scratch3,
                                                         Register scratch4,
                                                         Register scratch5,
                                                         Label* not_found) {
   // Register scratch3 is the general scratch register in this function.
   Register scratch = scratch3;
 
   // Make sure that both characters are not digits as such strings has a
   // different hash algorithm. Don't try to look for these in the symbol table.
   Label not_array_index;
   __ sub(scratch, c1, Operand(static_cast<int>('0')));
-  __ cmp(scratch, Operand(static_cast<int>('9' - '0')));
-  __ b(hi, &not_array_index);
+  __ cmphi(scratch, Operand(static_cast<int>('9' - '0')));
+  __ bt_near(&not_array_index);
   __ sub(scratch, c2, Operand(static_cast<int>('0')));
-  __ cmp(scratch, Operand(static_cast<int>('9' - '0')));
+  __ cmphi(scratch, Operand(static_cast<int>('9' - '0')));
 
   // If check failed combine both characters into single halfword.
   // This is required by the contract of the method: code at the
   // not_found branch expects this combination in c1 register
-  __ orr(c1, c1, Operand(c2, LSL, kBitsPerByte), LeaveCC, ls);
-  __ b(ls, not_found);
+  __ lsl(scratch1, c2, Operand(kBitsPerByte));
+  __ lor(c1, c1, scratch1, f);
+  __ b(f, not_found);
 
   __ bind(&not_array_index);
   // Calculate the two character string hash.
   Register hash = scratch1;
-  StringHelper::GenerateHashInit(masm, hash, c1);
-  StringHelper::GenerateHashAddCharacter(masm, hash, c2);
-  StringHelper::GenerateHashGetHash(masm, hash);
+  StringHelper::GenerateHashInit(masm, hash, c1, scratch);
+  StringHelper::GenerateHashAddCharacter(masm, hash, c2, scratch);
+  StringHelper::GenerateHashGetHash(masm, hash, scratch);
 
   // Collect the two characters in a register.
   Register chars = c1;
-  __ orr(chars, chars, Operand(c2, LSL, kBitsPerByte));
+  __ lsl(scratch, c2, Operand(kBitsPerByte));
+  __ lor(chars, chars, scratch);
 
   // chars: two character string, char 1 in byte 0 and char 2 in byte 1.
   // hash:  hash of two character string.
 
   // Load symbol table
   // Load address of first element of the symbol table.
   Register symbol_table = c2;
   __ LoadRoot(symbol_table, Heap::kSymbolTableRootIndex);
 
   Register undefined = scratch4;
   __ LoadRoot(undefined, Heap::kUndefinedValueRootIndex);
 
   // Calculate capacity mask from the symbol table capacity.
   Register mask = scratch2;
   __ ldr(mask, FieldMemOperand(symbol_table, SymbolTable::kCapacityOffset));
-  __ mov(mask, Operand(mask, ASR, 1));
+  __ asr(mask, mask, Operand(1));
   __ sub(mask, mask, Operand(1));
 
   // Calculate untagged address of the first element of the symbol table.
   Register first_symbol_table_element = symbol_table;
   __ add(first_symbol_table_element, symbol_table,
          Operand(SymbolTable::kElementsStartOffset - kHeapObjectTag));
 
   // Registers
   // chars: two character string, char 1 in byte 0 and char 2 in byte 1.
   // hash:  hash of two character string
   // mask:  capacity mask
   // first_symbol_table_element: address of the first element of
   //                             the symbol table
   // undefined: the undefined object
   // scratch: -
 
   // Perform a number of probes in the symbol table.
   const int kProbes = 4;
   Label found_in_symbol_table;
   Label next_probe[kProbes];
   Register candidate = scratch5;  // Scratch register contains candidate.
   for (int i = 0; i < kProbes; i++) {
     // Calculate entry in symbol table.
     if (i > 0) {
       __ add(candidate, hash, Operand(SymbolTable::GetProbeOffset(i)));
     } else {
       __ mov(candidate, hash);
     }
 
-    __ and_(candidate, candidate, Operand(mask));
+    __ land(candidate, candidate, mask);
 
     // Load the entry from the symble table.
     STATIC_ASSERT(SymbolTable::kEntrySize == 1);
-    __ ldr(candidate,
-           MemOperand(first_symbol_table_element,
-                      candidate,
-                      LSL,
-                      kPointerSizeLog2));
+    __ lsl(scratch, candidate, Operand(kPointerSizeLog2));
+    __ add(scratch, first_symbol_table_element, scratch);
+    __ ldr(candidate, MemOperand(scratch));
 
     // If entry is undefined no string with this hash can be found.
     Label is_string;
-    __ CompareObjectType(candidate, scratch, scratch, ODDBALL_TYPE);
-    __ b(ne, &is_string);
+    __ CompareObjectType(candidate, scratch, scratch, ODDBALL_TYPE, eq);
+    __ b(ne, &is_string, Label::kNear);
 
     __ cmp(undefined, candidate);
     __ b(eq, not_found);
     // Must be the hole (deleted entry).
     if (FLAG_debug_code) {
       __ LoadRoot(ip, Heap::kTheHoleValueRootIndex);
       __ cmp(ip, candidate);
       __ Assert(eq, "oddball in symbol table is not undefined or the hole");
     }
     __ jmp(&next_probe[i]);
@@ skipping 23 matching lines @@
 
   // Scratch register contains result when we fall through to here.
   Register result = candidate;
   __ bind(&found_in_symbol_table);
   __ Move(r0, result);
 }
 
 
 void StringHelper::GenerateHashInit(MacroAssembler* masm,
                                     Register hash,
-                                    Register character) {
+                                    Register character,
+                                    Register scratch) {
+  // Added a scratch parameter for the SH4 implementation compared to ARM.
   // hash = character + (character << 10);
   __ LoadRoot(hash, Heap::kHashSeedRootIndex);
   // Untag smi seed and add the character.
-  __ add(hash, character, Operand(hash, LSR, kSmiTagSize));
+  __ lsr(hash, hash, Operand(kSmiTagSize));
+  __ add(hash, character, hash);
   // hash += hash << 10;
-  __ add(hash, hash, Operand(hash, LSL, 10));
+  __ lsl(scratch, hash, Operand(10));
+  __ add(hash, hash, scratch);
   // hash ^= hash >> 6;
-  __ eor(hash, hash, Operand(hash, LSR, 6));
+  __ lsr(scratch, hash, Operand(6));
+  __ eor(hash, hash, scratch);
 }
 
 
 void StringHelper::GenerateHashAddCharacter(MacroAssembler* masm,
                                             Register hash,
-                                            Register character) {
+                                            Register character,
+                                            Register scratch) {
+  // Added a scratch parameter for the SH4 implementation compared to ARM.
   // hash += character;
-  __ add(hash, hash, Operand(character));
+  __ add(hash, hash, character);
   // hash += hash << 10;
-  __ add(hash, hash, Operand(hash, LSL, 10));
+  __ lsl(scratch, hash, Operand(10));
+  __ add(hash, hash, scratch);
   // hash ^= hash >> 6;
-  __ eor(hash, hash, Operand(hash, LSR, 6));
+  __ lsr(scratch, hash, Operand(6));
+  __ eor(hash, hash, scratch);
 }
 
 
 void StringHelper::GenerateHashGetHash(MacroAssembler* masm,
-                                       Register hash) {
+                                       Register hash,
+                                       Register scratch) {
+  // Added a scratch parameter for the SH4 implementation compared to ARM.
   // hash += hash << 3;
-  __ add(hash, hash, Operand(hash, LSL, 3));
+  __ lsl(scratch, hash, Operand(3));
+  __ add(hash, hash, scratch);
   // hash ^= hash >> 11;
-  __ eor(hash, hash, Operand(hash, LSR, 11));
+  __ lsr(scratch, hash, Operand(11));
+  __ eor(hash, hash, scratch);
   // hash += hash << 15;
-  __ add(hash, hash, Operand(hash, LSL, 15));
-
-  __ and_(hash, hash, Operand(String::kHashBitMask), SetCC);
+  __ lsl(scratch, hash, Operand(15));
+  __ add(hash, hash, scratch);
+  __ land(hash, hash, Operand(String::kHashBitMask));
+  __ cmpeq(hash, Operand(0));
 
   // if (hash == 0) hash = 27;
-  __ mov(hash, Operand(StringHasher::kZeroHash), LeaveCC, eq);
+  __ mov(hash, Operand(StringHasher::kZeroHash), eq);
 }
 
 
 void SubStringStub::Generate(MacroAssembler* masm) {
   Label runtime;
 
   // Stack frame on entry.
   //  lr: return address
   //  sp[0]: to
   //  sp[4]: from
   //  sp[8]: string
 
   // This stub is called from the native-call %_SubString(...), so
   // nothing can be assumed about the arguments. It is tested that:
   //  "string" is a sequential string,
   //  both "from" and "to" are smis, and
   //  0 <= from <= to <= string.length.
   // If any of these assumptions fail, we call the runtime system.
 
   const int kToOffset = 0 * kPointerSize;
   const int kFromOffset = 1 * kPointerSize;
   const int kStringOffset = 2 * kPointerSize;
 
   __ Ldrd(r2, r3, MemOperand(sp, kToOffset));
   STATIC_ASSERT(kFromOffset == kToOffset + 4);
   STATIC_ASSERT(kSmiTag == 0);
   STATIC_ASSERT(kSmiTagSize + kSmiShiftSize == 1);
+  __ JumpIfNotBothSmi(r2, r3, &runtime);      // not in arm code
+  // I.e., arithmetic shift right by one un-smi-tags.
+  __ asr(r2, r2, Operand(1));
+  __ asr(r3, r3, Operand(1));
+  __ cmpge(r3, Operand(0));
+  __ bf(&runtime);  // From is negative.
 
-  // I.e., arithmetic shift right by one un-smi-tags.
-  __ mov(r2, Operand(r2, ASR, 1), SetCC);
-  __ mov(r3, Operand(r3, ASR, 1), SetCC, cc);
-  // If either to or from had the smi tag bit set, then carry is set now.
-  __ b(cs, &runtime);  // Either "from" or "to" is not a smi.
-  // We want to bailout to runtime here if From is negative.  In that case, the
-  // next instruction is not executed and we fall through to bailing out to
-  // runtime.  pl is the opposite of mi.
-  // Both r2 and r3 are untagged integers.
-  __ sub(r2, r2, Operand(r3), SetCC, pl);
-  __ b(mi, &runtime);  // Fail if from > to.
+  // Both to and from are smis.
+  __ sub(r2, r2, r3);
+  __ cmpge(r2, Operand(0));
+  __ bf(&runtime);  // Fail if from > to.
 
   // Make sure first argument is a string.
   __ ldr(r0, MemOperand(sp, kStringOffset));
   STATIC_ASSERT(kSmiTag == 0);
   __ JumpIfSmi(r0, &runtime);
   Condition is_string = masm->IsObjectStringType(r0, r1);
   __ b(NegateCondition(is_string), &runtime);
 
   // Short-cut for the case of trivial substring.
   Label return_r0;
   // r0: original string
   // r2: result string length
   __ ldr(r4, FieldMemOperand(r0, String::kLengthOffset));
-  __ cmp(r2, Operand(r4, ASR, 1));
-  // Return original string.
-  __ b(eq, &return_r0);
+  __ asr(r4, r4, Operand(1));
+  __ cmpeq(r2, r4);
+  __ bt(&return_r0);
   // Longer than original string's length or negative: unsafe arguments.
-  __ b(hi, &runtime);
+  __ cmphi(r2, r4);
+  __ bt(&runtime);
   // Shorter than original string's length: an actual substring.
 
   // Deal with different string types: update the index if necessary
   // and put the underlying string into r5.
   // r0: original string
   // r1: instance type
   // r2: length
   // r3: from index (untagged)
   Label underlying_unpacked, sliced_string, seq_or_external_string;
   // If the string is not indirect, it can only be sequential or external.
@@ skipping 11 matching lines @@
   __ ldr(r5, FieldMemOperand(r0, ConsString::kFirstOffset));
   // Update instance type.
   __ ldr(r1, FieldMemOperand(r5, HeapObject::kMapOffset));
   __ ldrb(r1, FieldMemOperand(r1, Map::kInstanceTypeOffset));
   __ jmp(&underlying_unpacked);
 
   __ bind(&sliced_string);
   // Sliced string.  Fetch parent and correct start index by offset.
   __ ldr(r5, FieldMemOperand(r0, SlicedString::kParentOffset));
   __ ldr(r4, FieldMemOperand(r0, SlicedString::kOffsetOffset));
-  __ add(r3, r3, Operand(r4, ASR, 1));  // Add offset to index.
+  __ asr(r1, r4, Operand(1));
+  __ add(r3, r3, r1);  // Add offset to index.
   // Update instance type.
   __ ldr(r1, FieldMemOperand(r5, HeapObject::kMapOffset));
   __ ldrb(r1, FieldMemOperand(r1, Map::kInstanceTypeOffset));
   __ jmp(&underlying_unpacked);
 
   __ bind(&seq_or_external_string);
   // Sequential or external string.  Just move string to the expected register.
   __ mov(r5, r0);
 
   __ bind(&underlying_unpacked);
 
   if (FLAG_string_slices) {
     Label copy_routine;
     // r5: underlying subject string
     // r1: instance type of underlying subject string
     // r2: length
     // r3: adjusted start index (untagged)
-    __ cmp(r2, Operand(SlicedString::kMinLength));
+    __ cmpge(r2, Operand(SlicedString::kMinLength));
     // Short slice.  Copy instead of slicing.
-    __ b(lt, &copy_routine);
+    __ bf(&copy_routine);
     // Allocate new sliced string.  At this point we do not reload the instance
     // type including the string encoding because we simply rely on the info
     // provided by the original string.  It does not matter if the original
     // string's encoding is wrong because we always have to recheck encoding of
     // the newly created string's parent anyways due to externalized strings.
     Label two_byte_slice, set_slice_header;
     STATIC_ASSERT((kStringEncodingMask & kAsciiStringTag) != 0);
     STATIC_ASSERT((kStringEncodingMask & kTwoByteStringTag) == 0);
     __ tst(r1, Operand(kStringEncodingMask));
     __ b(eq, &two_byte_slice);
     __ AllocateAsciiSlicedString(r0, r2, r6, r7, &runtime);
     __ jmp(&set_slice_header);
     __ bind(&two_byte_slice);
     __ AllocateTwoByteSlicedString(r0, r2, r6, r7, &runtime);
     __ bind(&set_slice_header);
-    __ mov(r3, Operand(r3, LSL, 1));
+    __ lsl(r3, r3, Operand(1));
     __ str(r5, FieldMemOperand(r0, SlicedString::kParentOffset));
     __ str(r3, FieldMemOperand(r0, SlicedString::kOffsetOffset));
     __ jmp(&return_r0);
 
     __ bind(&copy_routine);
   }
 
   // r5: underlying subject string
   // r1: instance type of underlying subject string
   // r2: length
@@ skipping 40 matching lines @@
   StringHelper::GenerateCopyCharactersLong(masm, r1, r5, r2, r3, r4, r6, r7, r9,
                                            COPY_ASCII | DEST_ALWAYS_ALIGNED);
   __ jmp(&return_r0);
 
   // Allocate and copy the resulting two-byte string.
   __ bind(&two_byte_sequential);
   __ AllocateTwoByteString(r0, r2, r4, r6, r7, &runtime);
 
   // Locate first character of substring to copy.
   STATIC_ASSERT(kSmiTagSize == 1 && kSmiTag == 0);
-  __ add(r5, r5, Operand(r3, LSL, 1));
+  __ lsl(r1, r3, Operand(1));
+  __ add(r5, r5, r1);
   // Locate first character of result.
   __ add(r1, r0, Operand(SeqTwoByteString::kHeaderSize - kHeapObjectTag));
 
   // r0: result string.
   // r1: first character of result.
   // r2: result length.
   // r5: first character of substring to copy.
   STATIC_ASSERT((SeqTwoByteString::kHeaderSize & kObjectAlignmentMask) == 0);
   StringHelper::GenerateCopyCharactersLong(
       masm, r1, r5, r2, r3, r4, r6, r7, r9, DEST_ALWAYS_ALIGNED);
@@ skipping 16 matching lines @@
                                                       Register scratch1,
                                                       Register scratch2,
                                                       Register scratch3) {
   Register length = scratch1;
 
   // Compare lengths.
   Label strings_not_equal, check_zero_length;
   __ ldr(length, FieldMemOperand(left, String::kLengthOffset));
   __ ldr(scratch2, FieldMemOperand(right, String::kLengthOffset));
   __ cmp(length, scratch2);
-  __ b(eq, &check_zero_length);
+  __ b(eq, &check_zero_length, Label::kNear);
   __ bind(&strings_not_equal);
   __ mov(r0, Operand(Smi::FromInt(NOT_EQUAL)));
   __ Ret();
 
   // Check if the length is zero.
   Label compare_chars;
   __ bind(&check_zero_length);
   STATIC_ASSERT(kSmiTag == 0);
-  __ cmp(length, Operand(0));
-  __ b(ne, &compare_chars);
+  __ cmpeq(length, Operand(0));
+  __ b(ne, &compare_chars, Label::kNear);
   __ mov(r0, Operand(Smi::FromInt(EQUAL)));
   __ Ret();
 
   // Compare characters.
   __ bind(&compare_chars);
   GenerateAsciiCharsCompareLoop(masm,
                                 left, right, length, scratch2, scratch3,
                                 &strings_not_equal);
 
   // Characters are equal.
   __ mov(r0, Operand(Smi::FromInt(EQUAL)));
   __ Ret();
 }
 
 
 void StringCompareStub::GenerateCompareFlatAsciiStrings(MacroAssembler* masm,
                                                         Register left,
                                                         Register right,
                                                         Register scratch1,
                                                         Register scratch2,
                                                         Register scratch3,
                                                         Register scratch4) {
-  Label result_not_equal, compare_lengths;
+  ASSERT(!scratch2.is(r0) && !scratch4.is(r0));
+  Label result_not_equal, compare_lengths, skip;
   // Find minimum length and length difference.
   __ ldr(scratch1, FieldMemOperand(left, String::kLengthOffset));
   __ ldr(scratch2, FieldMemOperand(right, String::kLengthOffset));
-  __ sub(scratch3, scratch1, Operand(scratch2), SetCC);
+  __ cmpgt(scratch1, scratch2);  // for cond mov below
+  __ sub(scratch3, scratch1, scratch2);
   Register length_delta = scratch3;
-  __ mov(scratch1, scratch2, LeaveCC, gt);
+  __ mov(scratch1, scratch2, t);
   Register min_length = scratch1;
   STATIC_ASSERT(kSmiTag == 0);
-  __ cmp(min_length, Operand(0));
+  __ cmpeq(min_length, Operand(0));
   __ b(eq, &compare_lengths);
 
   // Compare loop.
   GenerateAsciiCharsCompareLoop(masm,
                                 left, right, min_length, scratch2, scratch4,
                                 &result_not_equal);
 
   // Compare lengths - strings up to min-length are equal.
   __ bind(&compare_lengths);
   ASSERT(Smi::FromInt(EQUAL) == static_cast<Smi*>(0));
   // Use length_delta as result if it's zero.
-  __ mov(r0, Operand(length_delta), SetCC);
+  __ mov(r0, length_delta);
+  __ cmpgt(r0, Operand(0));
+  __ mov(r0, Operand(Smi::FromInt(GREATER)), t);
+  __ cmpge(r0, Operand(0));
+  __ mov(r0, Operand(Smi::FromInt(LESS)), f);
+  __ Ret();
+
   __ bind(&result_not_equal);
   // Conditionally update the result based either on length_delta or
   // the last comparion performed in the loop above.
-  __ mov(r0, Operand(Smi::FromInt(GREATER)), LeaveCC, gt);
-  __ mov(r0, Operand(Smi::FromInt(LESS)), LeaveCC, lt);
+  __ cmpgt(scratch2, scratch4);
+  __ mov(r0, Operand(Smi::FromInt(GREATER)), t);
+  __ cmpge(scratch2, scratch4);
+  __ mov(r0, Operand(Smi::FromInt(LESS)), f);
   __ Ret();
 }
 
 
 void StringCompareStub::GenerateAsciiCharsCompareLoop(
     MacroAssembler* masm,
     Register left,
     Register right,
     Register length,
     Register scratch1,
     Register scratch2,
     Label* chars_not_equal) {
   // Change index to run from -length to -1 by adding length to string
   // start. This means that loop ends when index reaches zero, which
   // doesn't need an additional compare.
   __ SmiUntag(length);
   __ add(scratch1, length,
          Operand(SeqAsciiString::kHeaderSize - kHeapObjectTag));
-  __ add(left, left, Operand(scratch1));
-  __ add(right, right, Operand(scratch1));
+  __ add(left, left, scratch1);
+  __ add(right, right, scratch1);
   __ rsb(length, length, Operand::Zero());
   Register index = length;  // index = -length;
 
   // Compare loop.
   Label loop;
   __ bind(&loop);
   __ ldrb(scratch1, MemOperand(left, index));
   __ ldrb(scratch2, MemOperand(right, index));
   __ cmp(scratch1, scratch2);
   __ b(ne, chars_not_equal);
-  __ add(index, index, Operand(1), SetCC);
+  __ add(index, index, Operand(1));
+  __ tst(index, index);
   __ b(ne, &loop);
 }
 
 
 void StringCompareStub::Generate(MacroAssembler* masm) {
   Label runtime;
 
   Counters* counters = masm->isolate()->counters();
 
   // Stack frame on entry.
   //  sp[0]: right string
   //  sp[4]: left string
   __ Ldrd(r0 , r1, MemOperand(sp));  // Load right in r0, left in r1.
 
   Label not_same;
   __ cmp(r0, r1);
-  __ b(ne, &not_same);
+  __ b(ne, &not_same, Label::kNear);
   STATIC_ASSERT(EQUAL == 0);
   STATIC_ASSERT(kSmiTag == 0);
   __ mov(r0, Operand(Smi::FromInt(EQUAL)));
   __ IncrementCounter(counters->string_compare_native(), 1, r1, r2);
   __ add(sp, sp, Operand(2 * kPointerSize));
   __ Ret();
 
   __ bind(&not_same);
 
   // Check that both objects are sequential ASCII strings.
@@ skipping 29 matching lines @@
   if (flags_ == NO_STRING_ADD_FLAGS) {
     __ JumpIfEitherSmi(r0, r1, &call_runtime);
     // Load instance types.
     __ ldr(r4, FieldMemOperand(r0, HeapObject::kMapOffset));
     __ ldr(r5, FieldMemOperand(r1, HeapObject::kMapOffset));
     __ ldrb(r4, FieldMemOperand(r4, Map::kInstanceTypeOffset));
     __ ldrb(r5, FieldMemOperand(r5, Map::kInstanceTypeOffset));
     STATIC_ASSERT(kStringTag == 0);
     // If either is not a string, go to runtime.
     __ tst(r4, Operand(kIsNotStringMask));
-    __ tst(r5, Operand(kIsNotStringMask), eq);
+    __ b(ne, &call_runtime);
+    __ tst(r5, Operand(kIsNotStringMask));
     __ b(ne, &call_runtime);
   } else {
     // Here at least one of the arguments is definitely a string.
     // We convert the one that is not known to be a string.
     if ((flags_ & NO_STRING_CHECK_LEFT_IN_STUB) == 0) {
       ASSERT((flags_ & NO_STRING_CHECK_RIGHT_IN_STUB) != 0);
       GenerateConvertArgument(
           masm, 1 * kPointerSize, r0, r2, r3, r4, r5, &call_builtin);
       builtin_id = Builtins::STRING_ADD_RIGHT;
     } else if ((flags_ & NO_STRING_CHECK_RIGHT_IN_STUB) == 0) {
       ASSERT((flags_ & NO_STRING_CHECK_LEFT_IN_STUB) != 0);
       GenerateConvertArgument(
           masm, 0 * kPointerSize, r1, r2, r3, r4, r5, &call_builtin);
       builtin_id = Builtins::STRING_ADD_LEFT;
     }
   }
 
   // Both arguments are strings.
   // r0: first string
   // r1: second string
   // r4: first string instance type (if flags_ == NO_STRING_ADD_FLAGS)
   // r5: second string instance type (if flags_ == NO_STRING_ADD_FLAGS)
   {
-    Label strings_not_empty;
+    Label strings_not_empty, string_return;
     // Check if either of the strings are empty. In that case return the other.
     __ ldr(r2, FieldMemOperand(r0, String::kLengthOffset));
     __ ldr(r3, FieldMemOperand(r1, String::kLengthOffset));
     STATIC_ASSERT(kSmiTag == 0);
     __ cmp(r2, Operand(Smi::FromInt(0)));  // Test if first string is empty.
-    __ mov(r0, Operand(r1), LeaveCC, eq);  // If first is empty, return second.
+    __ mov(r0, r1, eq);  // If first is empty, return second.
+    __ bt_near(&string_return);
     STATIC_ASSERT(kSmiTag == 0);
      // Else test if second string is empty.
-    __ cmp(r3, Operand(Smi::FromInt(0)), ne);
-    __ b(ne, &strings_not_empty);  // If either string was empty, return r0.
+    __ cmp(r3, Operand(Smi::FromInt(0)));
+    // If either string was empty, return r0.
+    __ b(ne, &strings_not_empty, Label::kNear);
 
+    __ bind(&string_return);
     __ IncrementCounter(counters->string_add_native(), 1, r2, r3);
     __ add(sp, sp, Operand(2 * kPointerSize));
     __ Ret();
 
     __ bind(&strings_not_empty);
   }
 
-  __ mov(r2, Operand(r2, ASR, kSmiTagSize));
-  __ mov(r3, Operand(r3, ASR, kSmiTagSize));
+  __ asr(r2, r2, Operand(kSmiTagSize));
+  __ asr(r3, r3, Operand(kSmiTagSize));
   // Both strings are non-empty.
   // r0: first string
   // r1: second string
   // r2: length of first string
   // r3: length of second string
   // r4: first string instance type (if flags_ == NO_STRING_ADD_FLAGS)
   // r5: second string instance type (if flags_ == NO_STRING_ADD_FLAGS)
   // Look at the length of the result of adding the two strings.
   Label string_add_flat_result, longer_than_two;
   // Adding two lengths can't overflow.
   STATIC_ASSERT(String::kMaxLength < String::kMaxLength * 2);
-  __ add(r6, r2, Operand(r3));
+  __ add(r6, r2, r3);
   // Use the symbol table when adding two one character strings, as it
   // helps later optimizations to return a symbol here.
   __ cmp(r6, Operand(2));
   __ b(ne, &longer_than_two);
 
   // Check that both strings are non-external ASCII strings.
   if (flags_ != NO_STRING_ADD_FLAGS) {
     __ ldr(r4, FieldMemOperand(r0, HeapObject::kMapOffset));
     __ ldr(r5, FieldMemOperand(r1, HeapObject::kMapOffset));
     __ ldrb(r4, FieldMemOperand(r4, Map::kInstanceTypeOffset));
@@ skipping 23 matching lines @@
   // in a little endian mode)
   __ mov(r6, Operand(2));
   __ AllocateAsciiString(r0, r6, r4, r5, r9, &call_runtime);
   __ strh(r2, FieldMemOperand(r0, SeqAsciiString::kHeaderSize));
   __ IncrementCounter(counters->string_add_native(), 1, r2, r3);
   __ add(sp, sp, Operand(2 * kPointerSize));
   __ Ret();
 
   __ bind(&longer_than_two);
   // Check if resulting string will be flat.
-  __ cmp(r6, Operand(ConsString::kMinLength));
-  __ b(lt, &string_add_flat_result);
+  __ cmpge(r6, Operand(ConsString::kMinLength));
+  __ bf(&string_add_flat_result);
   // Handle exceptionally long strings in the runtime system.
   STATIC_ASSERT((String::kMaxLength & 0x80000000) == 0);
   ASSERT(IsPowerOf2(String::kMaxLength + 1));
   // kMaxLength + 1 is representable as shifted literal, kMaxLength is not.
-  __ cmp(r6, Operand(String::kMaxLength + 1));
-  __ b(hs, &call_runtime);
+  __ cmphs(r6, Operand(String::kMaxLength + 1));
+  __ bt(&call_runtime);
 
   // If result is not supposed to be flat, allocate a cons string object.
   // If both strings are ASCII the result is an ASCII cons string.
   if (flags_ != NO_STRING_ADD_FLAGS) {
     __ ldr(r4, FieldMemOperand(r0, HeapObject::kMapOffset));
     __ ldr(r5, FieldMemOperand(r1, HeapObject::kMapOffset));
     __ ldrb(r4, FieldMemOperand(r4, Map::kInstanceTypeOffset));
     __ ldrb(r5, FieldMemOperand(r5, Map::kInstanceTypeOffset));
   }
   Label non_ascii, allocated, ascii_data;
   STATIC_ASSERT(kTwoByteStringTag == 0);
   __ tst(r4, Operand(kStringEncodingMask));
-  __ tst(r5, Operand(kStringEncodingMask), ne);
-  __ b(eq, &non_ascii);
+  __ bt_near(&non_ascii);
+  __ tst(r5, Operand(kStringEncodingMask));
+  __ b(eq, &non_ascii, Label::kNear);
 
   // Allocate an ASCII cons string.
   __ bind(&ascii_data);
   __ AllocateAsciiConsString(r7, r6, r4, r5, &call_runtime);
   __ bind(&allocated);
   // Fill the fields of the cons string.
   __ str(r0, FieldMemOperand(r7, ConsString::kFirstOffset));
   __ str(r1, FieldMemOperand(r7, ConsString::kSecondOffset));
-  __ mov(r0, Operand(r7));
+  __ mov(r0, r7);
   __ IncrementCounter(counters->string_add_native(), 1, r2, r3);
   __ add(sp, sp, Operand(2 * kPointerSize));
   __ Ret();
 
   __ bind(&non_ascii);
   // At least one of the strings is two-byte. Check whether it happens
   // to contain only ASCII characters.
   // r4: first instance type.
   // r5: second instance type.
   __ tst(r4, Operand(kAsciiDataHintMask));
-  __ tst(r5, Operand(kAsciiDataHintMask), ne);
   __ b(ne, &ascii_data);
-  __ eor(r4, r4, Operand(r5));
+  __ tst(r5, Operand(kAsciiDataHintMask));
+  __ b(ne, &ascii_data);
+  __ eor(r4, r4, r5);
   STATIC_ASSERT(kAsciiStringTag != 0 && kAsciiDataHintTag != 0);
-  __ and_(r4, r4, Operand(kAsciiStringTag | kAsciiDataHintTag));
+  __ land(r4, r4, Operand(kAsciiStringTag | kAsciiDataHintTag));
   __ cmp(r4, Operand(kAsciiStringTag | kAsciiDataHintTag));
   __ b(eq, &ascii_data);
 
   // Allocate a two byte cons string.
   __ AllocateTwoByteConsString(r7, r6, r4, r5, &call_runtime);
   __ jmp(&allocated);
 
   // We cannot encounter sliced strings or cons strings here since:
   STATIC_ASSERT(SlicedString::kMinLength >= ConsString::kMinLength);
   // Handle creating a flat result from either external or sequential strings.
   // Locate the first characters' locations.
   // r0: first string
   // r1: second string
   // r2: length of first string
   // r3: length of second string
   // r4: first string instance type (if flags_ == NO_STRING_ADD_FLAGS)
   // r5: second string instance type (if flags_ == NO_STRING_ADD_FLAGS)
   // r6: sum of lengths.
   Label first_prepared, second_prepared;
   __ bind(&string_add_flat_result);
   if (flags_ != NO_STRING_ADD_FLAGS) {
     __ ldr(r4, FieldMemOperand(r0, HeapObject::kMapOffset));
     __ ldr(r5, FieldMemOperand(r1, HeapObject::kMapOffset));
     __ ldrb(r4, FieldMemOperand(r4, Map::kInstanceTypeOffset));
     __ ldrb(r5, FieldMemOperand(r5, Map::kInstanceTypeOffset));
   }
 
   // Check whether both strings have same encoding
-  __ eor(r7, r4, Operand(r5));
+  Label skip;
+  __ eor(r7, r4, r5);
   __ tst(r7, Operand(kStringEncodingMask));
   __ b(ne, &call_runtime);
 
   STATIC_ASSERT(kSeqStringTag == 0);
   __ tst(r4, Operand(kStringRepresentationMask));
+  __ bf(&skip);
   STATIC_ASSERT(SeqAsciiString::kHeaderSize == SeqTwoByteString::kHeaderSize);
   __ add(r7,
          r0,
-         Operand(SeqAsciiString::kHeaderSize - kHeapObjectTag),
-         LeaveCC,
-         eq);
-  __ b(eq, &first_prepared);
+         Operand(SeqAsciiString::kHeaderSize - kHeapObjectTag));
+  __ b(&first_prepared);
+  __ bind(&skip);
   // External string: rule out short external string and load string resource.
   STATIC_ASSERT(kShortExternalStringTag != 0);
   __ tst(r4, Operand(kShortExternalStringMask));
   __ b(ne, &call_runtime);
   __ ldr(r7, FieldMemOperand(r0, ExternalString::kResourceDataOffset));
   __ bind(&first_prepared);
 
   STATIC_ASSERT(kSeqStringTag == 0);
+  Label skip2;
   __ tst(r5, Operand(kStringRepresentationMask));
+  __ bf(&skip2);
   STATIC_ASSERT(SeqAsciiString::kHeaderSize == SeqTwoByteString::kHeaderSize);
   __ add(r1,
          r1,
-         Operand(SeqAsciiString::kHeaderSize - kHeapObjectTag),
-         LeaveCC,
-         eq);
-  __ b(eq, &second_prepared);
+         Operand(SeqAsciiString::kHeaderSize - kHeapObjectTag));
+  __ b(&second_prepared);
+  __ bind(&skip2);
   // External string: rule out short external string and load string resource.
   STATIC_ASSERT(kShortExternalStringTag != 0);
   __ tst(r5, Operand(kShortExternalStringMask));
   __ b(ne, &call_runtime);
   __ ldr(r1, FieldMemOperand(r1, ExternalString::kResourceDataOffset));
   __ bind(&second_prepared);
 
   Label non_ascii_string_add_flat_result;
   // r7: first character of first string
   // r1: first character of second string
@@ skipping 51 matching lines @@
                                             int stack_offset,
                                             Register arg,
                                             Register scratch1,
                                             Register scratch2,
                                             Register scratch3,
                                             Register scratch4,
                                             Label* slow) {
   // First check if the argument is already a string.
   Label not_string, done;
   __ JumpIfSmi(arg, &not_string);
-  __ CompareObjectType(arg, scratch1, scratch1, FIRST_NONSTRING_TYPE);
-  __ b(lt, &done);
+  __ CompareObjectType(arg, scratch1, scratch1, FIRST_NONSTRING_TYPE, ge);
+  __ bf(&done);
 
   // Check the number to string cache.
   Label not_cached;
   __ bind(&not_string);
   // Puts the cached result into scratch1.
   NumberToStringStub::GenerateLookupNumberStringCache(masm,
                                                       arg,
                                                       scratch1,
                                                       scratch2,
                                                       scratch3,
                                                       scratch4,
                                                       false,
                                                       &not_cached);
   __ mov(arg, scratch1);
   __ str(arg, MemOperand(sp, stack_offset));
   __ jmp(&done);
 
   // Check if the argument is a safe string wrapper.
   __ bind(&not_cached);
   __ JumpIfSmi(arg, slow);
   __ CompareObjectType(
-      arg, scratch1, scratch2, JS_VALUE_TYPE);  // map -> scratch1.
+      arg, scratch1, scratch2, JS_VALUE_TYPE, eq);  // map -> scratch1.
   __ b(ne, slow);
   __ ldrb(scratch2, FieldMemOperand(scratch1, Map::kBitField2Offset));
-  __ and_(scratch2,
+  __ land(scratch2,
           scratch2, Operand(1 << Map::kStringWrapperSafeForDefaultValueOf));
   __ cmp(scratch2,
          Operand(1 << Map::kStringWrapperSafeForDefaultValueOf));
   __ b(ne, slow);
   __ ldr(arg, FieldMemOperand(arg, JSValue::kValueOffset));
   __ str(arg, MemOperand(sp, stack_offset));
 
   __ bind(&done);
 }
 
 
 void ICCompareStub::GenerateSmis(MacroAssembler* masm) {
   ASSERT(state_ == CompareIC::SMIS);
   Label miss;
   __ orr(r2, r1, r0);
   __ JumpIfNotSmi(r2, &miss);
 
   if (GetCondition() == eq) {
     // For equality we do not care about the sign of the result.
-    __ sub(r0, r0, r1, SetCC);
+    __ sub(r0, r0, r1);
+    __ tst(r0, r0);     // TODO(stm): why setting CC? is it used?
   } else {
     // Untag before subtracting to avoid handling overflow.
     __ SmiUntag(r1);
-    __ sub(r0, r1, SmiUntagOperand(r0));
+    __ SmiUntag(r0);
+    __ sub(r0, r1, r0);
   }
   __ Ret();
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
 void ICCompareStub::GenerateHeapNumbers(MacroAssembler* masm) {
   ASSERT(state_ == CompareIC::HEAP_NUMBERS);
 
   Label generic_stub;
   Label unordered, maybe_undefined1, maybe_undefined2;
   Label miss;
-  __ and_(r2, r1, Operand(r0));
-  __ JumpIfSmi(r2, &generic_stub);
+  __ land(r2, r1, r0);
+  __ JumpIfSmi(r2, &generic_stub, Label::kNear);
 
-  __ CompareObjectType(r0, r2, r2, HEAP_NUMBER_TYPE);
-  __ b(ne, &maybe_undefined1);
-  __ CompareObjectType(r1, r2, r2, HEAP_NUMBER_TYPE);
-  __ b(ne, &maybe_undefined2);
+  __ CompareObjectType(r0, r2, r2, HEAP_NUMBER_TYPE, eq);
+  __ b(ne, &maybe_undefined1, Label::kNear);
+  __ CompareObjectType(r1, r2, r2, HEAP_NUMBER_TYPE, eq);
+  __ b(ne, &maybe_undefined2, Label::kNear);
 
   // Inlining the double comparison and falling back to the general compare
-  // stub if NaN is involved or VFP3 is unsupported.
-  if (CpuFeatures::IsSupported(VFP2)) {
-    CpuFeatures::Scope scope(VFP2);
+  // stub if NaN is involved or FPU is unsupported.
+  if (CpuFeatures::IsSupported(FPU)) {
+    // Load left and right operand
+    __ sub(r2, r1, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dldr(dr0, MemOperand(r2, 0), r2);
+    __ sub(r2, r0, Operand(kHeapObjectTag - HeapNumber::kValueOffset));
+    __ dldr(dr2, MemOperand(r2, 0), r2);
 
-    // Load left and right operand
-    __ sub(r2, r1, Operand(kHeapObjectTag));
-    __ vldr(d0, r2, HeapNumber::kValueOffset);
-    __ sub(r2, r0, Operand(kHeapObjectTag));
-    __ vldr(d1, r2, HeapNumber::kValueOffset);
+    Label unordered;
+    __ dcmpeq(dr0, dr0);
+    __ bf_near(&unordered);
+    __ dcmpeq(dr2, dr2);
+    __ bf_near(&unordered);
 
-    // Compare operands
-    __ VFPCompareAndSetFlags(d0, d1);
+    // Test for eq, lt and gt
+    Label equal, greater;
+    __ dcmpeq(dr0, dr2);
+    __ bt_near(&equal);
+    __ dcmpgt(dr0, dr2);
+    __ bt_near(&greater);
 
-    // Don't base result on status bits when a NaN is involved.
-    __ b(vs, &unordered);
+    __ mov(r0, Operand(LESS));
+    __ rts();
 
-    // Return a result of -1, 0, or 1, based on status bits.
-    __ mov(r0, Operand(EQUAL), LeaveCC, eq);
-    __ mov(r0, Operand(LESS), LeaveCC, lt);
-    __ mov(r0, Operand(GREATER), LeaveCC, gt);
-    __ Ret();
+    __ bind(&equal);
+    __ mov(r0, Operand(EQUAL));
+    __ rts();
+
+    __ bind(&greater);
+    __ mov(r0, Operand(GREATER));
+    __ rts();
+
+    __ bind(&unordered);
   }
 
-  __ bind(&unordered);
   CompareStub stub(GetCondition(), strict(), NO_COMPARE_FLAGS, r1, r0);
   __ bind(&generic_stub);
   __ Jump(stub.GetCode(), RelocInfo::CODE_TARGET);
 
   __ bind(&maybe_undefined1);
   if (Token::IsOrderedRelationalCompareOp(op_)) {
     __ CompareRoot(r0, Heap::kUndefinedValueRootIndex);
     __ b(ne, &miss);
-    __ CompareObjectType(r1, r2, r2, HEAP_NUMBER_TYPE);
+    __ CompareObjectType(r1, r2, r2, HEAP_NUMBER_TYPE, eq);
     __ b(ne, &maybe_undefined2);
     __ jmp(&unordered);
   }
 
   __ bind(&maybe_undefined2);
   if (Token::IsOrderedRelationalCompareOp(op_)) {
     __ CompareRoot(r1, Heap::kUndefinedValueRootIndex);
     __ b(eq, &unordered);
   }
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
 void ICCompareStub::GenerateSymbols(MacroAssembler* masm) {
   ASSERT(state_ == CompareIC::SYMBOLS);
   Label miss;
 
   // Registers containing left and right operands respectively.
   Register left = r1;
   Register right = r0;
   Register tmp1 = r2;
   Register tmp2 = r3;
 
   // Check that both operands are heap objects.
-  __ JumpIfEitherSmi(left, right, &miss);
+  __ JumpIfEitherSmi(left, right, &miss, Label::kNear);
 
   // Check that both operands are symbols.
   __ ldr(tmp1, FieldMemOperand(left, HeapObject::kMapOffset));
   __ ldr(tmp2, FieldMemOperand(right, HeapObject::kMapOffset));
   __ ldrb(tmp1, FieldMemOperand(tmp1, Map::kInstanceTypeOffset));
   __ ldrb(tmp2, FieldMemOperand(tmp2, Map::kInstanceTypeOffset));
   STATIC_ASSERT(kSymbolTag != 0);
-  __ and_(tmp1, tmp1, Operand(tmp2));
+  __ land(tmp1, tmp1, tmp2);
   __ tst(tmp1, Operand(kIsSymbolMask));
-  __ b(eq, &miss);
+  __ b(eq, &miss, Label::kNear);
 
   // Symbols are compared by identity.
   __ cmp(left, right);
   // Make sure r0 is non-zero. At this point input operands are
   // guaranteed to be non-zero.
   ASSERT(right.is(r0));
   STATIC_ASSERT(EQUAL == 0);
   STATIC_ASSERT(kSmiTag == 0);
-  __ mov(r0, Operand(Smi::FromInt(EQUAL)), LeaveCC, eq);
+  __ mov(r0, Operand(Smi::FromInt(EQUAL)), eq);
   __ Ret();
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
 void ICCompareStub::GenerateStrings(MacroAssembler* masm) {
   ASSERT(state_ == CompareIC::STRINGS);
   Label miss;
@@ skipping 19 matching lines @@
   __ ldrb(tmp2, FieldMemOperand(tmp2, Map::kInstanceTypeOffset));
   STATIC_ASSERT(kNotStringTag != 0);
   __ orr(tmp3, tmp1, tmp2);
   __ tst(tmp3, Operand(kIsNotStringMask));
   __ b(ne, &miss);
 
   // Fast check for identical strings.
   __ cmp(left, right);
   STATIC_ASSERT(EQUAL == 0);
   STATIC_ASSERT(kSmiTag == 0);
-  __ mov(r0, Operand(Smi::FromInt(EQUAL)), LeaveCC, eq);
+  __ mov(r0, Operand(Smi::FromInt(EQUAL)), eq);
   __ Ret(eq);
 
   // Handle not identical strings.
 
   // Check that both strings are symbols. If they are, we're done
   // because we already know they are not identical.
   if (equality) {
     ASSERT(GetCondition() == eq);
     STATIC_ASSERT(kSymbolTag != 0);
-    __ and_(tmp3, tmp1, Operand(tmp2));
+    __ land(tmp3, tmp1, tmp2);
     __ tst(tmp3, Operand(kIsSymbolMask));
     // Make sure r0 is non-zero. At this point input operands are
     // guaranteed to be non-zero.
     ASSERT(right.is(r0));
     __ Ret(ne);
   }
 
   // Check that both strings are sequential ASCII.
   Label runtime;
   __ JumpIfBothInstanceTypesAreNotSequentialAscii(
@@ skipping 18 matching lines @@
   }
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
 void ICCompareStub::GenerateObjects(MacroAssembler* masm) {
   ASSERT(state_ == CompareIC::OBJECTS);
   Label miss;
-  __ and_(r2, r1, Operand(r0));
-  __ JumpIfSmi(r2, &miss);
+  __ land(r2, r1, r0);
+  __ JumpIfSmi(r2, &miss, Label::kNear);
 
-  __ CompareObjectType(r0, r2, r2, JS_OBJECT_TYPE);
-  __ b(ne, &miss);
-  __ CompareObjectType(r1, r2, r2, JS_OBJECT_TYPE);
-  __ b(ne, &miss);
+  __ CompareObjectType(r0, r2, r2, JS_OBJECT_TYPE, eq);
+  __ b(ne, &miss, Label::kNear);
+  __ CompareObjectType(r1, r2, r2, JS_OBJECT_TYPE, eq);
+  __ b(ne, &miss, Label::kNear);
 
   ASSERT(GetCondition() == eq);
-  __ sub(r0, r0, Operand(r1));
+  __ sub(r0, r0, r1);
   __ Ret();
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
 void ICCompareStub::GenerateKnownObjects(MacroAssembler* masm) {
   Label miss;
-  __ and_(r2, r1, Operand(r0));
+  __ land(r2, r1, r0);
   __ JumpIfSmi(r2, &miss);
   __ ldr(r2, FieldMemOperand(r0, HeapObject::kMapOffset));
   __ ldr(r3, FieldMemOperand(r1, HeapObject::kMapOffset));
   __ cmp(r2, Operand(known_map_));
   __ b(ne, &miss);
   __ cmp(r3, Operand(known_map_));
   __ b(ne, &miss);
 
-  __ sub(r0, r0, Operand(r1));
+  __ sub(r0, r0, r1);
   __ Ret();
 
   __ bind(&miss);
   GenerateMiss(masm);
 }
 
 
-
 void ICCompareStub::GenerateMiss(MacroAssembler* masm) {
   {
     // Call the runtime system in a fresh internal frame.
     ExternalReference miss =
         ExternalReference(IC_Utility(IC::kCompareIC_Miss), masm->isolate());
 
     FrameScope scope(masm, StackFrame::INTERNAL);
     __ Push(r1, r0);
-    __ push(lr);
+    __ push(pr);
     __ Push(r1, r0);
     __ mov(ip, Operand(Smi::FromInt(op_)));
     __ push(ip);
     __ CallExternalReference(miss, 3);
     // Compute the entry point of the rewritten stub.
     __ add(r2, r0, Operand(Code::kHeaderSize - kHeapObjectTag));
     // Restore registers.
-    __ pop(lr);
+    __ pop(pr);
     __ pop(r0);
     __ pop(r1);
   }
 
-  __ Jump(r2);
+  __ jmp(r2);
 }
 
 
 void DirectCEntryStub::Generate(MacroAssembler* masm) {
-  __ ldr(pc, MemOperand(sp, 0));
+  __ ldr(scratch_, MemOperand(sp, 0), scratch_);
+  __ jmp(scratch_);
 }
 
 
 void DirectCEntryStub::GenerateCall(MacroAssembler* masm,
-                                    ExternalReference function) {
-  __ mov(r2, Operand(function));
-  GenerateCall(masm, r2);
+                                    ExternalReference function,
+                                    Register scratch1,
+                                    Register scratch2) {
+  __ mov(scratch1, Operand(function));
+  GenerateCall(masm, scratch1, scratch2);
 }
 
 
 void DirectCEntryStub::GenerateCall(MacroAssembler* masm,
-                                    Register target) {
-  __ mov(lr, Operand(reinterpret_cast<intptr_t>(GetCode().location()),
-                     RelocInfo::CODE_TARGET));
-
-  // Prevent literal pool emission during calculation of return address.
-  Assembler::BlockConstPoolScope block_const_pool(masm);
-
+                                    Register target,
+                                    Register scratch1) {
+  ASSERT(!target.is(scratch_));
+  ASSERT(!target.is(scratch1));
+  ASSERT(!scratch1.is(scratch_));
+  // Get pr (pointing to DirectCEntryStub::Generate) into scratch1
+  // pr can't be used directly as it is clobbered by addpc later
+  __ mov(scratch1, Operand(reinterpret_cast<intptr_t>(GetCode().location()),
+                   RelocInfo::CODE_TARGET));
+  int return_address_offset = 4 * Assembler::kInstrSize;
+  Label start;
   // Push return address (accessible to GC through exit frame pc).
-  // Note that using pc with str is deprecated.
-  Label start;
+  __ addpc(scratch_, return_address_offset, pr);
   __ bind(&start);
-  __ add(ip, pc, Operand(Assembler::kInstrSize));
-  __ str(ip, MemOperand(sp, 0));
-  __ Jump(target);  // Call the C++ function.
-  ASSERT_EQ(Assembler::kInstrSize + Assembler::kPcLoadDelta,
+  // restore the right pr (pointing to DirectCEntryStub::Generate)
+  __ mov(pr, scratch1);
+  __ str(scratch_, MemOperand(sp, 0), no_reg);
+  __ jmp(target);   // Call the api function.
+  ASSERT_EQ(return_address_offset,
             masm->SizeOfCodeGeneratedSince(&start));
 }
 
 
 void StringDictionaryLookupStub::GenerateNegativeLookup(MacroAssembler* masm,
                                                         Label* miss,
                                                         Label* done,
                                                         Register receiver,
                                                         Register properties,
                                                         Handle<String> name,
                                                         Register scratch0) {
+  ASSERT(!scratch0.is(ip));
   // If names of slots in range from 1 to kProbes - 1 for the hash value are
   // not equal to the name and kProbes-th slot is not used (its name is the
   // undefined value), it guarantees the hash table doesn't contain the
   // property. It's true even if some slots represent deleted properties
   // (their names are the hole value).
   for (int i = 0; i < kInlinedProbes; i++) {
     // scratch0 points to properties hash.
     // Compute the masked index: (hash + i + i * i) & mask.
     Register index = scratch0;
     // Capacity is smi 2^n.
     __ ldr(index, FieldMemOperand(properties, kCapacityOffset));
     __ sub(index, index, Operand(1));
-    __ and_(index, index, Operand(
+    __ land(index, index, Operand(
         Smi::FromInt(name->Hash() + StringDictionary::GetProbeOffset(i))));
 
     // Scale the index by multiplying by the entry size.
     ASSERT(StringDictionary::kEntrySize == 3);
-    __ add(index, index, Operand(index, LSL, 1));  // index *= 3.
+    __ lsl(ip, index, Operand(1));
+    __ add(index, index, ip);  // index *= 3.
 
     Register entity_name = scratch0;
     // Having undefined at this place means the name is not contained.
     ASSERT_EQ(kSmiTagSize, 1);
     Register tmp = properties;
-    __ add(tmp, properties, Operand(index, LSL, 1));
+    /* use entity_name as scratch (defined just after) */
+    __ lsl(entity_name, index, Operand(1));
+    __ add(tmp, properties, entity_name);
     __ ldr(entity_name, FieldMemOperand(tmp, kElementsStartOffset));
 
     ASSERT(!tmp.is(entity_name));
     __ LoadRoot(tmp, Heap::kUndefinedValueRootIndex);
     __ cmp(entity_name, tmp);
     __ b(eq, done);
 
     if (i != kInlinedProbes - 1) {
       // Load the hole ready for use below:
       __ LoadRoot(tmp, Heap::kTheHoleValueRootIndex);
@@ skipping 15 matching lines @@
 
       __ bind(&the_hole);
 
       // Restore the properties.
       __ ldr(properties,
              FieldMemOperand(receiver, JSObject::kPropertiesOffset));
     }
   }
 
   const int spill_mask =
-      (lr.bit() | r6.bit() | r5.bit() | r4.bit() | r3.bit() |
+      (r6.bit() | r5.bit() | r4.bit() | r3.bit() |
        r2.bit() | r1.bit() | r0.bit());
 
-  __ stm(db_w, sp, spill_mask);
+  __ push(pr);
+  __ pushm(spill_mask);
   __ ldr(r0, FieldMemOperand(receiver, JSObject::kPropertiesOffset));
   __ mov(r1, Operand(Handle<String>(name)));
   StringDictionaryLookupStub stub(NEGATIVE_LOOKUP);
   __ CallStub(&stub);
   __ cmp(r0, Operand(0));
-  __ ldm(ia_w, sp, spill_mask);
+  __ popm(spill_mask);
+  __ pop(pr);
 
   __ b(eq, done);
   __ b(ne, miss);
 }
 
 
 // Probe the string dictionary in the |elements| register. Jump to the
 // |done| label if a property with the given name is found. Jump to
 // the |miss| label otherwise.
 // If lookup was successful |scratch2| will be equal to elements + 4 * index.
 void StringDictionaryLookupStub::GeneratePositiveLookup(MacroAssembler* masm,
                                                         Label* miss,
                                                         Label* done,
                                                         Register elements,
                                                         Register name,
                                                         Register scratch1,
                                                         Register scratch2) {
   ASSERT(!elements.is(scratch1));
   ASSERT(!elements.is(scratch2));
   ASSERT(!name.is(scratch1));
   ASSERT(!name.is(scratch2));
 
-  __ AssertString(name);
+  // Assert that name contains a string.
+  if (FLAG_debug_code) __ AbortIfNotString(name);
 
   // Compute the capacity mask.
   __ ldr(scratch1, FieldMemOperand(elements, kCapacityOffset));
-  __ mov(scratch1, Operand(scratch1, ASR, kSmiTagSize));  // convert smi to int
+  __ asr(scratch1, scratch1, Operand(kSmiTagSize));  // convert smi to int
   __ sub(scratch1, scratch1, Operand(1));
 
   // Generate an unrolled loop that performs a few probes before
   // giving up. Measurements done on Gmail indicate that 2 probes
   // cover ~93% of loads from dictionaries.
   for (int i = 0; i < kInlinedProbes; i++) {
     // Compute the masked index: (hash + i + i * i) & mask.
     __ ldr(scratch2, FieldMemOperand(name, String::kHashFieldOffset));
     if (i > 0) {
       // Add the probe offset (i + i * i) left shifted to avoid right shifting
       // the hash in a separate instruction. The value hash + i + i * i is right
       // shifted in the following and instruction.
       ASSERT(StringDictionary::GetProbeOffset(i) <
              1 << (32 - String::kHashFieldOffset));
       __ add(scratch2, scratch2, Operand(
           StringDictionary::GetProbeOffset(i) << String::kHashShift));
     }
-    __ and_(scratch2, scratch1, Operand(scratch2, LSR, String::kHashShift));
+    __ lsr(scratch2, scratch2, Operand(String::kHashShift));
+    __ land(scratch2, scratch1, scratch2);
 
     // Scale the index by multiplying by the element size.
     ASSERT(StringDictionary::kEntrySize == 3);
     // scratch2 = scratch2 * 3.
-    __ add(scratch2, scratch2, Operand(scratch2, LSL, 1));
+    __ lsl(ip, scratch2, Operand(1));
+    __ add(scratch2, scratch2, ip);
 
     // Check if the key is identical to the name.
-    __ add(scratch2, elements, Operand(scratch2, LSL, 2));
+    __ lsl(scratch2, scratch2, Operand(2));
+    __ add(scratch2, elements, scratch2);
     __ ldr(ip, FieldMemOperand(scratch2, kElementsStartOffset));
-    __ cmp(name, Operand(ip));
+    __ cmp(name, ip);
     __ b(eq, done);
   }
 
   const int spill_mask =
-      (lr.bit() | r6.bit() | r5.bit() | r4.bit() |
+      (r6.bit() | r5.bit() | r4.bit() |
        r3.bit() | r2.bit() | r1.bit() | r0.bit()) &
       ~(scratch1.bit() | scratch2.bit());
 
-  __ stm(db_w, sp, spill_mask);
+  __ push(pr);
+  __ pushm(spill_mask);
   if (name.is(r0)) {
     ASSERT(!elements.is(r1));
     __ Move(r1, name);
     __ Move(r0, elements);
   } else {
     __ Move(r0, elements);
     __ Move(r1, name);
   }
   StringDictionaryLookupStub stub(POSITIVE_LOOKUP);
   __ CallStub(&stub);
   __ cmp(r0, Operand(0));
-  __ mov(scratch2, Operand(r2));
-  __ ldm(ia_w, sp, spill_mask);
+  __ mov(scratch2, r2);
+  __ popm(spill_mask);
+  __ pop(pr);
 
   __ b(ne, done);
   __ b(eq, miss);
 }
 
 
 void StringDictionaryLookupStub::Generate(MacroAssembler* masm) {
   // This stub overrides SometimesSetsUpAFrame() to return false.  That means
   // we cannot call anything that could cause a GC from this stub.
   // Registers:
@@ skipping 10 matching lines @@
   Register key = r1;
   Register index = r2;
   Register mask = r3;
   Register hash = r4;
   Register undefined = r5;
   Register entry_key = r6;
 
   Label in_dictionary, maybe_in_dictionary, not_in_dictionary;
 
   __ ldr(mask, FieldMemOperand(dictionary, kCapacityOffset));
-  __ mov(mask, Operand(mask, ASR, kSmiTagSize));
+  __ asr(mask, mask, Operand(kSmiTagSize));
   __ sub(mask, mask, Operand(1));
 
   __ ldr(hash, FieldMemOperand(key, String::kHashFieldOffset));
 
   __ LoadRoot(undefined, Heap::kUndefinedValueRootIndex);
 
   for (int i = kInlinedProbes; i < kTotalProbes; i++) {
     // Compute the masked index: (hash + i + i * i) & mask.
     // Capacity is smi 2^n.
     if (i > 0) {
       // Add the probe offset (i + i * i) left shifted to avoid right shifting
       // the hash in a separate instruction. The value hash + i + i * i is right
       // shifted in the following and instruction.
       ASSERT(StringDictionary::GetProbeOffset(i) <
              1 << (32 - String::kHashFieldOffset));
       __ add(index, hash, Operand(
           StringDictionary::GetProbeOffset(i) << String::kHashShift));
     } else {
-      __ mov(index, Operand(hash));
+      __ mov(index, hash);
     }
-    __ and_(index, mask, Operand(index, LSR, String::kHashShift));
+    __ lsr(index, index, Operand(String::kHashShift));
+    __ land(index, mask, index);
 
     // Scale the index by multiplying by the entry size.
     ASSERT(StringDictionary::kEntrySize == 3);
-    __ add(index, index, Operand(index, LSL, 1));  // index *= 3.
+    __ lsl(ip, index, Operand(1));
+    __ add(index, index, ip);  // index *= 3.
 
     ASSERT_EQ(kSmiTagSize, 1);
-    __ add(index, dictionary, Operand(index, LSL, 2));
+    __ lsl(index, index, Operand(2));
+    __ add(index, dictionary, index);
     __ ldr(entry_key, FieldMemOperand(index, kElementsStartOffset));
 
     // Having undefined at this place means the name is not contained.
-    __ cmp(entry_key, Operand(undefined));
+    __ cmp(entry_key, undefined);
     __ b(eq, &not_in_dictionary);
 
     // Stop if found the property.
-    __ cmp(entry_key, Operand(key));
+    __ cmp(entry_key, key);
     __ b(eq, &in_dictionary);
 
     if (i != kTotalProbes - 1 && mode_ == NEGATIVE_LOOKUP) {
       // Check if the entry name is not a symbol.
       __ ldr(entry_key, FieldMemOperand(entry_key, HeapObject::kMapOffset));
       __ ldrb(entry_key,
               FieldMemOperand(entry_key, Map::kInstanceTypeOffset));
       __ tst(entry_key, Operand(kIsSymbolMask));
       __ b(eq, &maybe_in_dictionary);
     }
   }
 
   __ bind(&maybe_in_dictionary);
   // If we are doing negative lookup then probing failure should be
   // treated as a lookup success. For positive lookup probing failure
   // should be treated as lookup failure.
   if (mode_ == POSITIVE_LOOKUP) {
     __ mov(result, Operand::Zero());
     __ Ret();
   }
 
   __ bind(&in_dictionary);
   __ mov(result, Operand(1));
   __ Ret();
 
   __ bind(&not_in_dictionary);
-  __ mov(result, Operand::Zero());
+  __ mov(result, Operand(0));
   __ Ret();
 }
 
 
 struct AheadOfTimeWriteBarrierStubList {
   Register object, value, address;
   RememberedSetAction action;
 };
 
 #define REG(Name) { kRegister_ ## Name ## _Code }
@@ skipping 71 matching lines @@
                          entry->value,
                          entry->address,
                          entry->action,
                          kDontSaveFPRegs);
     stub.GetCode()->set_is_pregenerated(true);
   }
 }
 
 
 bool CodeStub::CanUseFPRegisters() {
-  return CpuFeatures::IsSupported(VFP2);
+  return CpuFeatures::IsSupported(FPU);
 }
 
 
 // Takes the input in 3 registers: address_ value_ and object_.  A pointer to
 // the value has just been written into the object, now this stub makes sure
 // we keep the GC informed.  The word in the object where the value has been
 // written is in the address register.
 void RecordWriteStub::Generate(MacroAssembler* masm) {
   Label skip_to_incremental_noncompacting;
   Label skip_to_incremental_compacting;
 
   // The first two instructions are generated with labels so as to get the
   // offset fixed up correctly by the bind(Label*) call.  We patch it back and
   // forth between a compare instructions (a nop in this position) and the
   // real branch when we start and stop incremental heap marking.
   // See RecordWriteStub::Patch for details.
   {
     // Block literal pool emission, as the position of these two instructions
     // is assumed by the patching code.
-    Assembler::BlockConstPoolScope block_const_pool(masm);
     __ b(&skip_to_incremental_noncompacting);
     __ b(&skip_to_incremental_compacting);
   }
 
   if (remembered_set_action_ == EMIT_REMEMBERED_SET) {
     __ RememberedSetHelper(object_,
                            address_,
                            value_,
                            save_fp_regs_mode_,
                            MacroAssembler::kReturnAtEnd);
   }
   __ Ret();
 
   __ bind(&skip_to_incremental_noncompacting);
   GenerateIncremental(masm, INCREMENTAL);
 
   __ bind(&skip_to_incremental_compacting);
   GenerateIncremental(masm, INCREMENTAL_COMPACTION);
 
   // Initial mode of the stub is expected to be STORE_BUFFER_ONLY.
   // Will be checked in IncrementalMarking::ActivateGeneratedStub.
-  ASSERT(Assembler::GetBranchOffset(masm->instr_at(0)) < (1 << 12));
-  ASSERT(Assembler::GetBranchOffset(masm->instr_at(4)) < (1 << 12));
-  PatchBranchIntoNop(masm, 0);
-  PatchBranchIntoNop(masm, Assembler::kInstrSize);
+// TODO(STM): to check soon !
+//  ASSERT(Assembler::GetBranchOffset(masm->instr_at(0)) < (1 << 12));
+//  ASSERT(Assembler::GetBranchOffset(masm->instr_at(4)) < (1 << 12));
+//  PatchBranchIntoNop(masm, 0);
+//  PatchBranchIntoNop(masm, Assembler::kInstrSize);
 }
 
 
 void RecordWriteStub::GenerateIncremental(MacroAssembler* masm, Mode mode) {
   regs_.Save(masm);
 
   if (remembered_set_action_ == EMIT_REMEMBERED_SET) {
     Label dont_need_remembered_set;
 
     __ ldr(regs_.scratch0(), MemOperand(regs_.address(), 0));
@@ skipping 24 matching lines @@
 
   CheckNeedsToInformIncrementalMarker(
       masm, kReturnOnNoNeedToInformIncrementalMarker, mode);
   InformIncrementalMarker(masm, mode);
   regs_.Restore(masm);
   __ Ret();
 }
 
 
 void RecordWriteStub::InformIncrementalMarker(MacroAssembler* masm, Mode mode) {
-  regs_.SaveCallerSaveRegisters(masm, save_fp_regs_mode_);
-  int argument_count = 3;
-  __ PrepareCallCFunction(argument_count, regs_.scratch0());
-  Register address =
-      r0.is(regs_.address()) ? regs_.scratch0() : regs_.address();
-  ASSERT(!address.is(regs_.object()));
-  ASSERT(!address.is(r0));
-  __ Move(address, regs_.address());
-  __ Move(r0, regs_.object());
-  if (mode == INCREMENTAL_COMPACTION) {
-    __ Move(r1, address);
-  } else {
-    ASSERT(mode == INCREMENTAL);
-    __ ldr(r1, MemOperand(address, 0));
-  }
-  __ mov(r2, Operand(ExternalReference::isolate_address()));
-
-  AllowExternalCallThatCantCauseGC scope(masm);
-  if (mode == INCREMENTAL_COMPACTION) {
-    __ CallCFunction(
-        ExternalReference::incremental_evacuation_record_write_function(
-            masm->isolate()),
-        argument_count);
-  } else {
-    ASSERT(mode == INCREMENTAL);
-    __ CallCFunction(
-        ExternalReference::incremental_marking_record_write_function(
-            masm->isolate()),
-        argument_count);
-  }
-  regs_.RestoreCallerSaveRegisters(masm, save_fp_regs_mode_);
+  __ UNIMPLEMENTED_BREAK();
 }
 
 
 void RecordWriteStub::CheckNeedsToInformIncrementalMarker(
     MacroAssembler* masm,
     OnNoNeedToInformIncrementalMarker on_no_need,
     Mode mode) {
   Label on_black;
   Label need_incremental;
   Label need_incremental_pop_scratch;
 
-  __ and_(regs_.scratch0(), regs_.object(), Operand(~Page::kPageAlignmentMask));
+  __ land(regs_.scratch0(), regs_.object(), Operand(~Page::kPageAlignmentMask));
   __ ldr(regs_.scratch1(),
          MemOperand(regs_.scratch0(),
                     MemoryChunk::kWriteBarrierCounterOffset));
-  __ sub(regs_.scratch1(), regs_.scratch1(), Operand(1), SetCC);
+  __ sub(regs_.scratch1(), regs_.scratch1(), Operand(1));
+  __ cmpge(regs_.scratch1(), Operand(0));
   __ str(regs_.scratch1(),
          MemOperand(regs_.scratch0(),
                     MemoryChunk::kWriteBarrierCounterOffset));
-  __ b(mi, &need_incremental);
+  __ bf(&need_incremental);
 
   // Let's look at the color of the object:  If it is not black we don't have
   // to inform the incremental marker.
   __ JumpIfBlack(regs_.object(), regs_.scratch0(), regs_.scratch1(), &on_black);
 
   regs_.Restore(masm);
   if (on_no_need == kUpdateRememberedSetOnNoNeedToInformIncrementalMarker) {
     __ RememberedSetHelper(object_,
                            address_,
                            value_,
@@ skipping 82 matching lines @@
   // call.
   __ Push(r1, r3, r0);
   __ ldr(r5, MemOperand(fp, JavaScriptFrameConstants::kFunctionOffset));
   __ ldr(r5, FieldMemOperand(r5, JSFunction::kLiteralsOffset));
   __ Push(r5, r4);
   __ TailCallRuntime(Runtime::kStoreArrayLiteralElement, 5, 1);
 
   // Array literal has ElementsKind of FAST_*_ELEMENTS and value is an object.
   __ bind(&fast_elements);
   __ ldr(r5, FieldMemOperand(r1, JSObject::kElementsOffset));
-  __ add(r6, r5, Operand(r3, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r6, r3, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r6, r5, r6);
   __ add(r6, r6, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
   __ str(r0, MemOperand(r6, 0));
   // Update the write barrier for the array store.
   __ RecordWrite(r5, r6, r0, kLRHasNotBeenSaved, kDontSaveFPRegs,
                  EMIT_REMEMBERED_SET, OMIT_SMI_CHECK);
   __ Ret();
 
   // Array literal has ElementsKind of FAST_*_SMI_ELEMENTS or FAST_*_ELEMENTS,
   // and value is Smi.
   __ bind(&smi_element);
   __ ldr(r5, FieldMemOperand(r1, JSObject::kElementsOffset));
-  __ add(r6, r5, Operand(r3, LSL, kPointerSizeLog2 - kSmiTagSize));
+  __ lsl(r6, r3, Operand(kPointerSizeLog2 - kSmiTagSize));
+  __ add(r6, r5, r6);
   __ str(r0, FieldMemOperand(r6, FixedArray::kHeaderSize));
   __ Ret();
 
   // Array literal has ElementsKind of FAST_DOUBLE_ELEMENTS.
   __ bind(&double_elements);
   __ ldr(r5, FieldMemOperand(r1, JSObject::kElementsOffset));
   __ StoreNumberToDoubleElements(r0, r3, r1,
                                  // Overwrites all regs after this.
                                  r5, r6, r7, r9, r2,
                                  &slow_elements);
   __ Ret();
 }
 
 
 void ProfileEntryHookStub::MaybeCallEntryHook(MacroAssembler* masm) {
   if (entry_hook_ != NULL) {
-    PredictableCodeSizeScope predictable(masm);
     ProfileEntryHookStub stub;
     __ push(lr);
     __ CallStub(&stub);
     __ pop(lr);
   }
 }
 
 
 void ProfileEntryHookStub::Generate(MacroAssembler* masm) {
   // The entry hook is a "push lr" instruction, followed by a call.
   const int32_t kReturnAddressDistanceFromFunctionStart =
-      3 * Assembler::kInstrSize;
+      Assembler::kCallTargetAddressOffset + Assembler::kInstrSize;
 
   // Save live volatile registers.
-  __ Push(lr, r5, r1);
+  __ Push(pr, r5, r1);
   const int32_t kNumSavedRegs = 3;
 
   // Compute the function's address for the first argument.
-  __ sub(r0, lr, Operand(kReturnAddressDistanceFromFunctionStart));
+  __ sub(r0, pr, Operand(kReturnAddressDistanceFromFunctionStart));
 
   // The caller's return address is above the saved temporaries.
   // Grab that for the second argument to the hook.
   __ add(r1, sp, Operand(kNumSavedRegs * kPointerSize));
 
   // Align the stack if necessary.
-  int frame_alignment = masm->ActivationFrameAlignment();
+  int frame_alignment = OS::ActivationFrameAlignment();
   if (frame_alignment > kPointerSize) {
     __ mov(r5, sp);
     ASSERT(IsPowerOf2(frame_alignment));
-    __ and_(sp, sp, Operand(-frame_alignment));
+    __ land(sp, sp, Operand(-frame_alignment));
   }
 
-#if defined(V8_HOST_ARCH_ARM)
   __ mov(ip, Operand(reinterpret_cast<int32_t>(&entry_hook_)));
   __ ldr(ip, MemOperand(ip));
-#else
-  // Under the simulator we need to indirect the entry hook through a
-  // trampoline function at a known address.
-  Address trampoline_address = reinterpret_cast<Address>(
-      reinterpret_cast<intptr_t>(EntryHookTrampoline));
-  ApiFunction dispatcher(trampoline_address);
-  __ mov(ip, Operand(ExternalReference(&dispatcher,
-                                       ExternalReference::BUILTIN_CALL,
-                                       masm->isolate())));
-#endif
-  __ Call(ip);
+
+  __ jsr(ip);
 
   // Restore the stack pointer if needed.
   if (frame_alignment > kPointerSize) {
     __ mov(sp, r5);
   }
 
-  __ Pop(lr, r5, r1);
+  __ Pop(pr, r5, r1);
   __ Ret();
 }
 
 #undef __
 
 } }  // namespace v8::internal
 
-#endif  // V8_TARGET_ARCH_ARM
+#endif  // V8_TARGET_ARCH_SH4