Add ldrd and strd instructions to the ARM port. This is a

commit for zhangk@codeaurora.org.  See
http://codereview.chromium.org/568029 and
http://codereview.chromium.org/2019003

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@4618 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

AUTHORS

@@ -18,6 +18,7 @@ Jan de Mooij <jandemooij@gmail.com>
 Jay Freeman <saurik@saurik.com>
 Joel Stanley <joel.stan@gmail.com>
 John Jozwiak <jjozwiak@codeaurora.org>
+Kun Zhang <zhangk@codeaurora.org>
 Matt Hanselman <mjhanselman@gmail.com>
 Martyn Capewell <martyn.capewell@arm.com>
 Paolo Giarrusso <p.giarrusso@gmail.com>

src/arm/assembler-arm.cc

@@ -1157,6 +1157,35 @@ void Assembler::ldrsh(Register dst, const MemOperand& src, Condition cond) {
 }
 
 
+void Assembler::ldrd(Register dst, const MemOperand& src, Condition cond) {
+  ASSERT(src.rm().is(no_reg));
+#ifdef CAN_USE_ARMV7_INSTRUCTIONS
+  addrmod3(cond | B7 | B6 | B4, dst, src);
+#else
+  ldr(dst, src, cond);
+  MemOperand src1(src);
+  src1.set_offset(src1.offset() + 4);
+  Register dst1(dst);
+  dst1.code_ = dst1.code_ + 1;
+  ldr(dst1, src1, cond);
+#endif
+}
+
+
+void Assembler::strd(Register src, const MemOperand& dst, Condition cond) {
+  ASSERT(dst.rm().is(no_reg));
+#ifdef CAN_USE_ARMV7_INSTRUCTIONS
+  addrmod3(cond | B7 | B6 | B5 | B4, src, dst);
+#else
+  str(src, dst, cond);
+  MemOperand dst1(dst);
+  dst1.set_offset(dst1.offset() + 4);
+  Register src1(src);
+  src1.code_ = src1.code_ + 1;
+  str(src1, dst1, cond);
+#endif
+}
+
 // Load/Store multiple instructions.
 void Assembler::ldm(BlockAddrMode am,
                     Register base,

src/arm/assembler-arm.h

@@ -448,6 +448,18 @@ class MemOperand BASE_EMBEDDED {
   explicit MemOperand(Register rn, Register rm,
                       ShiftOp shift_op, int shift_imm, AddrMode am = Offset);
 
+  void set_offset(int32_t offset) {
+      ASSERT(rm_.is(no_reg));
+      offset_ = offset;
+  }
+
+  uint32_t offset() {
+      ASSERT(rm_.is(no_reg));
+      return offset_;
+  }
+
+  Register rm() const {return rm_;}
+
  private:
   Register rn_;  // base
   Register rm_;  // register offset
@@ -755,6 +767,8 @@ class Assembler : public Malloced {
   void strh(Register src, const MemOperand& dst, Condition cond = al);
   void ldrsb(Register dst, const MemOperand& src, Condition cond = al);
   void ldrsh(Register dst, const MemOperand& src, Condition cond = al);
+  void ldrd(Register dst, const MemOperand& src, Condition cond = al);
+  void strd(Register src, const MemOperand& dst, Condition cond = al);
 
   // Load/Store multiple instructions
   void ldm(BlockAddrMode am, Register base, RegList dst, Condition cond = al);

src/arm/codegen-arm.cc

@@ -1486,8 +1486,7 @@ void CodeGenerator::CallApplyLazy(Expression* applicand,
   // Then process it as a normal function call.
   __ ldr(r0, MemOperand(sp, 3 * kPointerSize));
   __ ldr(r1, MemOperand(sp, 2 * kPointerSize));
-  __ str(r0, MemOperand(sp, 2 * kPointerSize));
-  __ str(r1, MemOperand(sp, 3 * kPointerSize));
+  __ strd(r0, MemOperand(sp, 2 * kPointerSize));
 
   CallFunctionStub call_function(2, NOT_IN_LOOP, NO_CALL_FUNCTION_FLAGS);
   frame_->CallStub(&call_function, 3);
@@ -2279,8 +2278,8 @@ void CodeGenerator::VisitForInStatement(ForInStatement* node) {
   node->break_target()->set_direction(JumpTarget::FORWARD_ONLY);
   node->continue_target()->set_direction(JumpTarget::FORWARD_ONLY);
 
-  __ ldr(r0, frame_->ElementAt(0));  // load the current count
-  __ ldr(r1, frame_->ElementAt(1));  // load the length
+  // Load the current count to r0, load the length to r1.
+  __ ldrd(r0, frame_->ElementAt(0));
   __ cmp(r0, r1);  // compare to the array length
   node->break_target()->Branch(hs);
 
@@ -6310,8 +6309,7 @@ static void EmitSmiNonsmiComparison(MacroAssembler* masm,
     ConvertToDoubleStub stub1(r3, r2, r7, r6);
     __ Call(stub1.GetCode(), RelocInfo::CODE_TARGET);
     // Load rhs to a double in r0, r1.
-    __ ldr(r1, FieldMemOperand(r0, HeapNumber::kValueOffset + kPointerSize));
-    __ ldr(r0, FieldMemOperand(r0, HeapNumber::kValueOffset));
+    __ ldrd(r0, FieldMemOperand(r0, HeapNumber::kValueOffset));
     __ pop(lr);
   }
 
@@ -6346,8 +6344,7 @@ static void EmitSmiNonsmiComparison(MacroAssembler* masm,
   } else {
     __ push(lr);
     // Load lhs to a double in r2, r3.
-    __ ldr(r3, FieldMemOperand(r1, HeapNumber::kValueOffset + kPointerSize));
-    __ ldr(r2, FieldMemOperand(r1, HeapNumber::kValueOffset));
+    __ ldrd(r2, FieldMemOperand(r1, HeapNumber::kValueOffset));
     // Convert rhs to a double in r0, r1.
     __ mov(r7, Operand(r0));
     ConvertToDoubleStub stub2(r1, r0, r7, r6);
@@ -6511,10 +6508,8 @@ static void EmitCheckForTwoHeapNumbers(MacroAssembler* masm,
     __ sub(r7, r1, Operand(kHeapObjectTag));
     __ vldr(d7, r7, HeapNumber::kValueOffset);
   } else {
-    __ ldr(r2, FieldMemOperand(r1, HeapNumber::kValueOffset));
-    __ ldr(r3, FieldMemOperand(r1, HeapNumber::kValueOffset + kPointerSize));
-    __ ldr(r1, FieldMemOperand(r0, HeapNumber::kValueOffset + kPointerSize));
-    __ ldr(r0, FieldMemOperand(r0, HeapNumber::kValueOffset));
+    __ ldrd(r2, FieldMemOperand(r1, HeapNumber::kValueOffset));
+    __ ldrd(r0, FieldMemOperand(r0, HeapNumber::kValueOffset));
   }
   __ jmp(both_loaded_as_doubles);
 }
@@ -6891,8 +6886,7 @@ void GenericBinaryOpStub::HandleBinaryOpSlowCases(
       __ vldr(d7, r7, HeapNumber::kValueOffset);
     } else {
       // Calling convention says that second double is in r2 and r3.
-      __ ldr(r2, FieldMemOperand(r0, HeapNumber::kValueOffset));
-      __ ldr(r3, FieldMemOperand(r0, HeapNumber::kValueOffset + 4));
+      __ ldrd(r2, FieldMemOperand(r0, HeapNumber::kValueOffset));
     }
     __ jmp(&finished_loading_r0);
     __ bind(&r0_is_smi);
@@ -6944,8 +6938,7 @@ void GenericBinaryOpStub::HandleBinaryOpSlowCases(
       __ vldr(d6, r7, HeapNumber::kValueOffset);
     } else {
       // Calling convention says that first double is in r0 and r1.
-      __ ldr(r0, FieldMemOperand(r1, HeapNumber::kValueOffset));
-      __ ldr(r1, FieldMemOperand(r1, HeapNumber::kValueOffset + 4));
+      __ ldrd(r0, FieldMemOperand(r1, HeapNumber::kValueOffset));
     }
     __ jmp(&finished_loading_r1);
     __ bind(&r1_is_smi);
@@ -7016,8 +7009,7 @@ void GenericBinaryOpStub::HandleBinaryOpSlowCases(
       __ stc(p1, cr8, MemOperand(r4, HeapNumber::kValueOffset));
   #else
       // Double returned in registers 0 and 1.
-      __ str(r0, FieldMemOperand(r5, HeapNumber::kValueOffset));
-      __ str(r1, FieldMemOperand(r5, HeapNumber::kValueOffset + 4));
+      __ strd(r0, FieldMemOperand(r5, HeapNumber::kValueOffset));
   #endif
       __ mov(r0, Operand(r5));
       // And we are done.

src/arm/disasm-arm.cc

@@ -418,6 +418,12 @@ int Decoder::FormatOption(Instr* instr, const char* format) {
         ASSERT(STRING_STARTS_WITH(format, "memop"));
         if (instr->HasL()) {
           Print("ldr");
+        } else if ((instr->Bits(27, 25) == 0) && (instr->Bit(20) == 0)) {
+          if (instr->Bits(7, 4) == 0xf) {
+            Print("strd");
+          } else {
+            Print("ldrd");
+          }
         } else {
           Print("str");
         }
@@ -614,6 +620,47 @@ void Decoder::DecodeType01(Instr* instr) {
       } else {
         Unknown(instr);  // not used by V8
       }
+    } else if ((instr->Bit(20) == 0) && ((instr->Bits(7, 4) & 0xd) == 0xd)) {
+      // ldrd, strd
+      switch (instr->PUField()) {
+        case 0: {
+          if (instr->Bit(22) == 0) {
+            Format(instr, "'memop'cond's 'rd, ['rn], -'rm");
+          } else {
+            Format(instr, "'memop'cond's 'rd, ['rn], #-'off8");
+          }
+          break;
+        }
+        case 1: {
+          if (instr->Bit(22) == 0) {
+            Format(instr, "'memop'cond's 'rd, ['rn], +'rm");
+          } else {
+            Format(instr, "'memop'cond's 'rd, ['rn], #+'off8");
+          }
+          break;
+        }
+        case 2: {
+          if (instr->Bit(22) == 0) {
+            Format(instr, "'memop'cond's 'rd, ['rn, -'rm]'w");
+          } else {
+            Format(instr, "'memop'cond's 'rd, ['rn, #-'off8]'w");
+          }
+          break;
+        }
+        case 3: {
+          if (instr->Bit(22) == 0) {
+            Format(instr, "'memop'cond's 'rd, ['rn, +'rm]'w");
+          } else {
+            Format(instr, "'memop'cond's 'rd, ['rn, #+'off8]'w");
+          }
+          break;
+        }
+        default: {
+          // The PU field is a 2-bit field.
+          UNREACHABLE();
+          break;
+        }
+      }
     } else {
       // extra load/store instructions
       switch (instr->PUField()) {

src/arm/simulator-arm.cc

@@ -728,6 +728,13 @@ int32_t Simulator::get_register(int reg) const {
 }
 
 
+void Simulator::set_dw_register(int dreg, const int* dbl) {
+  ASSERT((dreg >= 0) && (dreg < num_d_registers));
+  registers_[dreg] = dbl[0];
+  registers_[dreg + 1] = dbl[1];
+}
+
+
 // Raw access to the PC register.
 void Simulator::set_pc(int32_t value) {
   pc_modified_ = true;
@@ -887,7 +894,7 @@ int Simulator::ReadW(int32_t addr, Instr* instr) {
     intptr_t* ptr = reinterpret_cast<intptr_t*>(addr);
     return *ptr;
   }
-  PrintF("Unaligned read at 0x%08x\n", addr);
+  PrintF("Unaligned read at 0x%08x, pc=%p\n", addr, instr);
   UNIMPLEMENTED();
   return 0;
 #endif
@@ -1001,6 +1008,41 @@ void Simulator::WriteB(int32_t addr, int8_t value) {
 }
 
 
+int32_t* Simulator::ReadDW(int32_t addr) {
+#if V8_TARGET_CAN_READ_UNALIGNED
+  int32_t* ptr = reinterpret_cast<int32_t*>(addr);
+  return ptr;
+#else
+  if ((addr & 3) == 0) {
+    int32_t* ptr = reinterpret_cast<int32_t*>(addr);
+    return ptr;
+  }
+  PrintF("Unaligned read at 0x%08x\n", addr);
+  UNIMPLEMENTED();
+  return 0;
+#endif
+}
+
+
+void Simulator::WriteDW(int32_t addr, int32_t value1, int32_t value2) {
+#if V8_TARGET_CAN_READ_UNALIGNED
+  int32_t* ptr = reinterpret_cast<int32_t*>(addr);
+  *ptr++ = value1;
+  *ptr = value2;
+  return;
+#else
+  if ((addr & 3) == 0) {
+    int32_t* ptr = reinterpret_cast<int32_t*>(addr);
+    *ptr++ = value1;
+    *ptr = value2;
+    return;
+  }
+  PrintF("Unaligned write at 0x%08x\n", addr);
+  UNIMPLEMENTED();
+#endif
+}
+
+
 // Returns the limit of the stack area to enable checking for stack overflows.
 uintptr_t Simulator::StackLimit() const {
   // Leave a safety margin of 256 bytes to prevent overrunning the stack when
@@ -1628,7 +1670,19 @@ void Simulator::DecodeType01(Instr* instr) {
           }
         }
       }
-      if (instr->HasH()) {
+      if (((instr->Bits(7, 4) & 0xd) == 0xd) && (instr->Bit(20) == 0)) {
+        ASSERT((rd % 2) == 0);
+        if (instr->HasH()) {
+          // The strd instruction.
+          int32_t value1 = get_register(rd);
+          int32_t value2 = get_register(rd+1);
+          WriteDW(addr, value1, value2);
+        } else {
+          // The ldrd instruction.
+          int* rn_data = ReadDW(addr);
+          set_dw_register(rd, rn_data);
+        }
+      } else if (instr->HasH()) {
         if (instr->HasSign()) {
           if (instr->HasL()) {
             int16_t val = ReadH(addr, instr);

src/arm/simulator-arm.h

@@ -159,6 +159,7 @@ class Simulator {
   // instruction.
   void set_register(int reg, int32_t value);
   int32_t get_register(int reg) const;
+  void set_dw_register(int dreg, const int* dbl);
 
   // Support for VFP.
   void set_s_register(int reg, unsigned int value);
@@ -252,6 +253,9 @@ class Simulator {
   inline int ReadW(int32_t addr, Instr* instr);
   inline void WriteW(int32_t addr, int value, Instr* instr);
 
+  int32_t* ReadDW(int32_t addr);
+  void WriteDW(int32_t addr, int32_t value1, int32_t value2);
+
   // Executing is handled based on the instruction type.
   void DecodeType01(Instr* instr);  // both type 0 and type 1 rolled into one
   void DecodeType2(Instr* instr);