Use custom dispatcher for Array.prototype routines

[rerobika]

Binary size reduction:

• Intel: ~2.5KB (gcc-7.3)
• Arm: ~1.2KB (arm-linux-gnueabi-gcc-7.3)

JerryScript-DCO-1.0-Signed-off-by: Robert Fancsik frobert@inf.u-szeged.hu

[rerobika]

Performance results:

Benchmark	Perf (sec)
3d-cube.js	0.790 -> 0.788 : +0.224%
3d-raytrace.js	0.950 -> 0.954 : -0.359%
access-binary-trees.js	0.528 -> 0.528 : +0.106%
access-fannkuch.js	1.913 -> 1.914 : -0.056%
access-nbody.js	1.000 -> 0.999 : +0.054%
bitops-3bit-bits-in-byte.js	0.470 -> 0.471 : -0.156%
bitops-bits-in-byte.js	0.645 -> 0.646 : -0.150%
bitops-bitwise-and.js	0.890 -> 0.890 : -0.061%
bitops-nsieve-bits.js	1.128 -> 1.125 : +0.258%
controlflow-recursive.js	0.333 -> 0.332 : +0.549%
crypto-aes.js	0.814 -> 0.809 : +0.644%
crypto-md5.js	0.555 -> 0.556 : -0.098%
crypto-sha1.js	0.542 -> 0.543 : -0.077%
date-format-tofte.js	0.726 -> 0.725 : +0.044%
date-format-xparb.js	0.508 -> 0.513 : -1.076%
math-cordic.js	1.104 -> 1.110 : -0.493%
math-partial-sums.js	0.704 -> 0.701 : +0.372%
math-spectral-norm.js	0.464 -> 0.465 : -0.303%
string-base64.js	1.409 -> 1.368 : +2.886%
string-fasta.js	1.307 -> 1.302 : +0.387%
Geometric mean:	+0.137%

Most of it is probably noise due to the new code layout.

Binary (bytes)	924f4bb	c1ab183	Gain
size	132664	132664	0.000%
.text	115596	114404	+1.031%
.rodata	11256	11262	-0.053%
.bss	1575168	1575168	0.000%

[zherczeg]

LGTM with minor fix.

[rerobika]

Note: #2640 must land first.

[LaszloLango]

@rerobika I don't see how is this PR related to #2640. Could you explain it to me?

[rerobika]

@LaszloLango #2640 Adds 4 new Array.prototype routines here, but currently it uses the general dispatcher. If it lands, this patch must be updated with these 4 functions custom dispatcher related bindings.

[LaszloLango]

@rerobika as I see it can be done the other way as well.

[rerobika]

Indeed, no matter what PR lands first, the other must be updated as well.

[rerobika]

@LaszloLango I've updated the PR after #2640 has landed. Could you take an other look at it?

[LaszloLango]

@rerobika, are measurement results the same after the rebase?

[rerobika]

@LaszloLango Yes they are, these are the new results:

Binary (bytes)	master(f22eea5)	patch(cec4788)	Gain
size	132664	132664	0 bytes
.text	115636	114444	-1192 bytes
.rodata	11256	11255	-1 bytes
.bss	1575168	1575168	0 bytes

Performance is still unchanged, mostly noise due to the new code layout.

[LaszloLango]

LGTM