ARM64: optimize TruncateDoubleToI

ARM64: optimize TruncateDoubleToI

R=ulan@chromium.org, bmeurer@chromium.org

BUG=

[vincent.belliard, 2014-06-27 15:09:24]

Use the same mechanism used by Push for pushing lr and one double.

Before patch:
sub csp, jssp, #0x8 (8)
str lr, [jssp, #-8]!
sub csp, jssp, #0x8 (8)
str d4, [jssp, #-8]!

After patch:
sub csp, jssp, #0x10 (16)
str d4, [jssp, #-16]!
str lr, [jssp, #8]

[ulan, 2014-06-30 12:51:53]

If this is performance neutral, I'd prefer not landing this because it reduces
readability.

[vincent.belliard, 2014-06-30 13:03:08]

It only reduces the code size by one instruction for each TruncateDoubleToI

[ulan, 2014-06-30 13:10:11]

An alternative would be to introduce Push(Register, DoubleRegister) which makes
this optimization and keeps TruncateDoubleToI simple.

[vincent.belliard, 2014-06-30 15:02:56]

done

[ulan, 2014-06-30 15:22:58]

Thanks, lgtm.