[SIMD] optimize the cvtColor kernel

[huningxin]

According to #300 (comment), the SIMD cvtColor kernel is times slower than scalar version.

According to my investigation, the root cause is the inefficient pshufb with memory operands are generated by V8 for current implementation.

One solution is to refer to sse implementation that uses punpcklbw and punpckhqdq. I tried but it still fails due to an emscripten issue that leads V8 fails to generate those instructions. Let's see the response from emscripten community.

[huningxin]

@Wenzhao-Xiang , please move this info to PR for record.

[huningxin]

Link to WebAssembly/simd#196

[huningxin]

@lionkunonly , please report the latest perf data you have. Thanks!

[lionkunonly]

Ubuntu 18.04, Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz, chrome dev 85.0.4183.26, and emscripten 1.39.16.

	WASM Scalar (mean time)	WASM SIMD (mean time)	SIMD speedup
CvtColor (1920x1080,COLOR_RGBA2GRAY)	2.4161	2.8215	0.86X

[lionkunonly]

@huningxin Hi, Ningxin. I am not sure how to optimize the CvtColor. Do you have more suggestions in detail?

[huningxin]

As mentioned in #301 (comment), this issue depends on WebAssembly/simd#196 and emscripten-core/emscripten#9340.

Compared to the last year's performance (0.24X), your latest data (0.86X) is 3.5X faster.

It seems WebAssembly community is still working on that. Let's monitor and test if there is new progress.