Dynamically align OSR frames on ia32.

src/ia32/lithium-codegen-ia32.cc

 // Copyright 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
@@ skipping 322 matching lines @@
 }
 
 
 void LCodeGen::GenerateOsrPrologue() {
   // Generate the OSR entry prologue at the first unknown OSR value, or if there
   // are none, at the OSR entrypoint instruction.
   if (osr_pc_offset_ >= 0) return;
 
   osr_pc_offset_ = masm()->pc_offset();
 
+    // Move state of dynamic frame alignment into edx.
+  __ mov(edx, Immediate(kNoAlignmentPadding));
+
+  if (support_aligned_spilled_doubles_ && dynamic_frame_alignment_) {
+    __ mov(ebx, ebp);
+    __ add(Operand(ebx), Immediate(kPointerSize));
+
+    Label do_not_pad, align_loop;
+    // Align ebp + 4 to a multiple of 2 * kPointerSize.
+    __ test(ebx, Immediate(kPointerSize));

[Yang, 2013/09/18 09:30:06]

Couldn't you test ebp directly? Something like

test(ebp, Immediate(kPointerSize));
j(zero, &do_not_pad)

+    __ j(not_zero, &do_not_pad, Label::kNear);
+    __ push(Immediate(0));
+    __ mov(ebx, esp);
+    __ mov(edx, Immediate(kAlignmentPaddingPushed));
+
+    // Move all parts of the frame over one word. The frame consists of:
+    // unoptimized frame slots, alignment state, context, frame pointer, return
+    // address, receiver, and the arguments.
+    __ mov(ecx, Immediate(scope()->num_parameters() +
+           5 + graph()->osr()->UnoptimizedFrameSlots()));
+
+    __ bind(&align_loop);
+    __ mov(eax, Operand(ebx, 1 * kPointerSize));
+    __ mov(Operand(ebx, 0), eax);
+    __ add(Operand(ebx), Immediate(kPointerSize));
+    __ dec(ecx);
+    __ j(not_zero, &align_loop, Label::kNear);
+    __ mov(Operand(ebx, 0), Immediate(kAlignmentZapValue));
+    __ sub(Operand(ebp), Immediate(kPointerSize));
+    __ bind(&do_not_pad);

[Yang, 2013/09/18 09:30:06]

You can probably save a instruction per loop if you use ecx as negative index
with ebp as base pointer, and increment ecx by a word in each loop until it hits
zero.

But this is easier to read, so I'm also fine with this.

+  }
+
   // Save the first local, which is overwritten by the alignment state.
   Operand alignment_loc = MemOperand(ebp, -3 * kPointerSize);
   __ push(alignment_loc);
 
-  // Set the dynamic frame alignment state to "not aligned".
-  __ mov(alignment_loc, Immediate(kNoAlignmentPadding));
+  // Set the dynamic frame alignment state.
+  __ mov(alignment_loc, edx);
 
   // Adjust the frame size, subsuming the unoptimized frame into the
   // optimized frame.
   int slots = GetStackSlotCount() - graph()->osr()->UnoptimizedFrameSlots();
   ASSERT(slots >= 1);
   __ sub(esp, Immediate((slots - 1) * kPointerSize));
 }
 
 
 bool LCodeGen::GenerateBody() {
@@ skipping 5991 matching lines @@
                               FixedArray::kHeaderSize - kPointerSize));
   __ bind(&done);
 }
 
 
 #undef __
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_IA32