Improving Parser Performance Using SSE Instructions

[kazuho]

http://blog.kazuhooku.com/2014/12/improving-parser-performance-using-sse.html

[rurban]

Using SSE4.2 or more is trivial. Same goes for fast hashes with the crc32 intrinsic. Mostly the compiler does it for you (-march=native)

But you cannot assume everybody has such a CPU. Hence you'd either need to compile your own (a la macports, gentoo, perl, ...) or do run-time checks for the CPU feature and switch to the fast version then.

[yuhong]

I wonder if there is any HTML parsers that use SSE4.2 instructions

[supersillyus → yuhong]

IIRC Servo is using html5ever for HTML parsing, which (at least in one branch) uses SSE4.2 for parsing.

[bmm6o]

It's an interesting idea, but I found this blog post unenlightening. Does anyone have links to other writeups?

[octo_t]

clang does a very similar thing in the preprocessor (looking for new lines if my memory serves correctly).

[billconan]

can http parsing be done on a GPU?

[michael_h → billconan]

Maybe, but the transfer time to/from the gpu is probably going to kill any gains you might get.