A64: Add and use a double register which holds the 0.0 value.

src/a64/codegen-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 528 matching lines @@
   //  NZCV
   //  1000      -708.4 < input < 709.8    result = exp(input)
   //  0110      input == 709.8            result = +infinity
   //  0010      input > 709.8             result = +infinity
   //  0011      input is NaN              result = input
   //  0000      input <= -708.4           result = +0.0
 
   // Continue the common case first. 'mi' tests N == 1.
   __ B(&result_is_finite_non_zero, mi);
 
-  // TODO(jbramley): Add (and use) a zero D register for A64.
   // TODO(jbramley): Consider adding a +infinity register for A64.
   __ Ldr(double_temp2, ExpConstant(constants, 2));    // Synthesize +infinity.
-  __ Fsub(double_temp1, double_temp1, double_temp1);  // Synthesize +0.0.
 
   // Select between +0.0 and +infinity. 'lo' tests C == 0.
-  __ Fcsel(result, double_temp1, double_temp2, lo);
+  __ Fcsel(result, fp_zero, double_temp2, lo);
   // Select between {+0.0 or +infinity} and input. 'vc' tests V == 0.
   __ Fcsel(result, result, input, vc);
   __ B(&done);
 
   // The rest is magic, as described in InitializeMathExpData().
   __ Bind(&result_is_finite_non_zero);
 
   // Assert that we can load offset 3 and offset 4 with a single ldp.
   ASSERT(kDRegSizeInBytes == (ExpConstant(constants, 4).offset() -
                               ExpConstant(constants, 3).offset()));
@@ skipping 42 matching lines @@
   __ Fmul(result, double_temp3, double_temp1);
 
   __ Bind(&done);
 }
 
 #undef __
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_A64