Four byte optimized memcpy.

[zherczeg]

Binary unchanged.

Benchmark	Perf (sec)
3d-cube.js	1.152 -> 1.147 : +0.378%
3d-raytrace.js	1.371 -> 1.315 : +4.092%
access-binary-trees.js	0.674 -> 0.673 : +0.062%
access-fannkuch.js	2.854 -> 2.845 : +0.315%
access-nbody.js	1.386 -> 1.387 : -0.047%
bitops-3bit-bits-in-byte.js	0.630 -> 0.631 : -0.154%
bitops-bits-in-byte.js	0.917 -> 0.917 : -0.063%
bitops-bitwise-and.js	1.920 -> 1.923 : -0.123%
bitops-nsieve-bits.js	2.276 -> 2.240 : +1.578%
controlflow-recursive.js	0.475 -> 0.475 : +0.026%
crypto-aes.js	1.251 -> 1.244 : +0.622%
crypto-md5.js	0.849 -> 0.845 : +0.473%
crypto-sha1.js	0.792 -> 0.789 : +0.432%
date-format-tofte.js	1.023 -> 1.019 : +0.409%
date-format-xparb.js	0.556 -> 0.551 : +0.820%
math-cordic.js	1.669 -> 1.667 : +0.100%
math-partial-sums.js	0.986 -> 0.986 : -0.013%
math-spectral-norm.js	0.705 -> 0.700 : +0.740%
string-base64.js	4.597 -> 2.241 : +51.246%
string-fasta.js	2.145 -> 2.124 : +1.000%
Geometric mean:	+4.046%

[LaszloLango]

LGTM

[dbatyai]

LGTM

[lemmaa]

@zherczeg , the return value for memcpy, s1, will change after apply your patch. Right? It seems that the behavior of original function also change. PTAL.

[zherczeg]

You are right. We don't use its return value, so we didn't have any issues. But I will fix it.