ARM: Make DoStoreKeyedFixedDoubleArray faster; don't allow conditional Vmov

ARM: Make DoStoreKeyedFixedDoubleArray faster; don't allow conditional Vmov

This patch makes us generate faster code for DoStoreKeyedFixedDoubleArray,
by using a branch rather than a conditional Vmov instruction.

Conditional VFP instructions are not a great idea in general, and it was
especially bad in this case because Vmov expands to a bunch of instructions.
For this reason, the patch also removes the 'cond' parameter from Vmov.

Thanks to Rodolph for pointing me to this!

BUG=none

Committed: https://code.google.com/p/v8/source/detail?r=13722

[hans, 2013-02-25 12:10:16]

This makes for faster writes to unboxed double arrays.

On the matmul benchmark that I used in
https://chromiumcodereview.appspot.com/11428137, I get 17% speed-up on Nexus 10.
On kraken-imaging-desaturate-orig, I get about 5% speed-up.

[ulan, 2013-02-25 12:47:05]

The change looks good to me. We can land it after getting Rodolph's review.

[Rodolph Perfetta, 2013-02-25 15:01:58]

LGTM.

[ulan, 2013-02-25 16:15:14]

Thans, LGTM, landing.

[ulan, 2013-02-25 16:15:44]

Committed patchset #1 manually as r13722 (presubmit successful).