Fix arm64 deoptimization from double registers (reverts r20613).

This reverts "ARM64: Use pair memory access in deoptimizer entry", r20613. It does not really make sense to micro-optimize the deoptimizer as it is the ultra-slow path. Moreover, the original code was easier to read (in addition to being correct).

BUG=391313
LOG=N
R=ulan@chromium.org

Review URL: https://codereview.chromium.org/389583003

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@22360 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/arm64/deoptimizer-arm64.cc

@@ -110,42 +110,6 @@ void Deoptimizer::CopyDoubleRegisters(FrameDescription* output_frame) {
 }
 
 
-#define __ masm->
-
-static void CopyRegisterDumpToFrame(MacroAssembler* masm,
-                                    Register frame,
-                                    CPURegList reg_list,
-                                    Register scratch1,
-                                    Register scratch2,
-                                    int src_offset,
-                                    int dst_offset) {
-  int offset0, offset1;
-  CPURegList copy_to_input = reg_list;
-  int reg_count = reg_list.Count();
-  int reg_size = reg_list.RegisterSizeInBytes();
-  for (int i = 0; i < (reg_count / 2); i++) {
-    __ PeekPair(scratch1, scratch2, src_offset + (i * reg_size * 2));
-
-    offset0 = (copy_to_input.PopLowestIndex().code() * reg_size) + dst_offset;
-    offset1 = (copy_to_input.PopLowestIndex().code() * reg_size) + dst_offset;
-
-    if ((offset0 + reg_size) == offset1) {
-      // Registers are adjacent: store in pairs.
-      __ Stp(scratch1, scratch2, MemOperand(frame, offset0));
-    } else {
-      // Registers are not adjacent: store individually.
-      __ Str(scratch1, MemOperand(frame, offset0));
-      __ Str(scratch2, MemOperand(frame, offset1));
-    }
-  }
-  if ((reg_count & 1) != 0) {
-    __ Peek(scratch1, src_offset + (reg_count - 1) * reg_size);
-    offset0 = (copy_to_input.PopLowestIndex().code() * reg_size) + dst_offset;
-    __ Str(scratch1, MemOperand(frame, offset0));
-  }
-}
-
-#undef __
 
 #define __ masm()->
 
@@ -209,13 +173,23 @@ void Deoptimizer::EntryGenerator::Generate() {
   __ Ldr(x1, MemOperand(deoptimizer, Deoptimizer::input_offset()));
 
   // Copy core registers into the input frame.
-  CopyRegisterDumpToFrame(masm(), x1, saved_registers, x2, x4, 0,
-                          FrameDescription::registers_offset());
+  CPURegList copy_to_input = saved_registers;
+  for (int i = 0; i < saved_registers.Count(); i++) {
+    __ Peek(x2, i * kPointerSize);
+    CPURegister current_reg = copy_to_input.PopLowestIndex();
+    int offset = (current_reg.code() * kPointerSize) +
+        FrameDescription::registers_offset();
+    __ Str(x2, MemOperand(x1, offset));
+  }
 
   // Copy FP registers to the input frame.
-  CopyRegisterDumpToFrame(masm(), x1, saved_fp_registers, x2, x4,
-                          kFPRegistersOffset,
-                          FrameDescription::double_registers_offset());
+  for (int i = 0; i < saved_fp_registers.Count(); i++) {
+    int dst_offset = FrameDescription::double_registers_offset() +
+        (i * kDoubleSize);
+    int src_offset = kFPRegistersOffset + (i * kDoubleSize);
+    __ Peek(x2, src_offset);
+    __ Str(x2, MemOperand(x1, dst_offset));
+  }
 
   // Remove the bailout id and the saved registers from the stack.
   __ Drop(1 + (kSavedRegistersAreaSize / kXRegSize));

test/mjsunit/regress/regress-381313.js

+// Copyright 2014 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// Flags: --allow-natives-syntax
+
+var g = 0;
+
+function f(x, deopt) {
+  var a0 = x;
+  var a1 = 2 * x;
+  var a2 = 3 * x;
+  var a3 = 4 * x;
+  var a4 = 5 * x;
+  var a5 = 6 * x;
+  var a6 = 7 * x;
+  var a7 = 8 * x;
+  var a8 = 9 * x;
+  var a9 = 10 * x;
+  var a10 = 11 * x;
+  var a11 = 12 * x;
+  var a12 = 13 * x;
+  var a13 = 14 * x;
+  var a14 = 15 * x;
+  var a15 = 16 * x;
+  var a16 = 17 * x;
+  var a17 = 18 * x;
+  var a18 = 19 * x;
+  var a19 = 20 * x;
+
+  g = 1;
+
+  deopt + 0;
+
+  return a0 + a1 + a2 + a3 + a4 + a5 + a6 + a7 + a8 + a9 +
+         a10 + a11 + a12 + a13 + a14 + a15 + a16 + a17 + a18 + a19;
+}
+
+f(0.5, 0);
+f(0.5, 0);
+%OptimizeFunctionOnNextCall(f);
+print(f(0.5, ""));