A64: Add support for --optimize-for-size in PushMultipleTimes

src/a64/macro-assembler-a64.cc

 // Copyright 2013 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 654 matching lines @@
     // but if we keep it matching StackPointer, the simulator can detect memory
     // accesses in the now-free part of the stack.
     Mov(csp, StackPointer());
   }
 }
 
 
 void MacroAssembler::PushMultipleTimes(int count, Register src) {
   int size = src.SizeInBytes();
 
-  // TODO(all): Use a loop when optimizing for size.
-  TODO_UNIMPLEMENTED("PushMultipleTimes: Support --optimize-for-size.");
+  PrepareForPush(count, size);
 
-  PrepareForPush(count, size);
+  if (FLAG_optimize_for_size && count > 4) {

[Rodolph Perfetta (ARM), 2014/02/06 15:14:36]

The code below is 4 instructions long, PushMultipleTimes can handle 8 push in 4
instructions (using stp) so the break even point should be 8.

+    Label loop;
+    __ Mov(Tmp0(), count);
+    __ Bind(&loop);
+    PushHelper(1, size, src, NoReg, NoReg, NoReg);

[ulan, 2014/02/06 14:53:12]

Optimization: push 4 values at once (count / 4) times, and then set count =
(count % 4) and fall through to the old code.

[Rodolph Perfetta (ARM), 2014/02/06 15:14:36]

if count is even you can push two at a time with no extra memory cost.

[Rodolph Perfetta (ARM), 2014/02/06 15:14:36]

4 at a time will be faster but bigger, is that what we want?

On 2014/02/06 14:53:12, ulan wrote:

+    __ Subs(Tmp0(), Tmp0(), 1);
+    __ B(ne, &loop);
+    return;
+  }
+
   // Push up to four registers at a time if possible because if the current
   // stack pointer is csp and the register size is 32, registers must be pushed
   // in blocks of four in order to maintain the 16-byte alignment for csp.
   while (count >= 4) {
     PushHelper(4, size, src, src, src, src);
     count -= 4;
   }
   if (count >= 2) {
     PushHelper(2, size, src, src, NoReg, NoReg);
     count -= 2;
@@ skipping 4141 matching lines @@
   }
 }
 
 
 #undef __
 
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_A64