Improve get-value

[zherczeg]

Slight perf improvement. But 4% on access-fannkuch.js

Benchmark	Perf (sec)
3d-cube.js	1.210 -> 1.178 : +2.620%
3d-raytrace.js	1.468 -> 1.415 : +3.627%
access-binary-trees.js	0.727 -> 0.730 : -0.408%
access-fannkuch.js	2.980 -> 2.837 : +4.802%
access-nbody.js	1.481 -> 1.408 : +4.900%
bitops-3bit-bits-in-byte.js	0.645 -> 0.651 : -0.937%
bitops-bits-in-byte.js	0.944 -> 0.942 : +0.160%
bitops-bitwise-and.js	1.971 -> 1.999 : -1.376%
bitops-nsieve-bits.js	2.284 -> 2.275 : +0.397%
controlflow-recursive.js	0.495 -> 0.497 : -0.348%
crypto-aes.js	1.360 -> 1.303 : +4.137%
crypto-md5.js	0.865 -> 0.872 : -0.803%
crypto-sha1.js	0.810 -> 0.812 : -0.270%
date-format-tofte.js	1.098 -> 1.088 : +0.924%
date-format-xparb.js	0.569 -> 0.578 : -1.470%
math-cordic.js	1.726 -> 1.715 : +0.659%
math-partial-sums.js	1.009 -> 1.008 : +0.123%
math-spectral-norm.js	0.737 -> 0.714 : +3.046%
string-base64.js	4.611 -> 4.629 : -0.388%
string-fasta.js	2.190 -> 2.178 : +0.565%
Geometric mean:	+1.019%

Binary size unchanged.

[LaszloLango]

That do { ... } while (0) style is bit odd to me, but nice improvement. If we cannot use a more common if-else style here with similar performance results, then it will be good in this form. LGTM

[zherczeg]

Patch updated.

[LaszloLango]

@zherczeg, thanks for the updated. LGTM

[dbatyai]

LGTM