[PATCH] D17293: [MC] AsmLexer: 30% speedup on tests, added extensible identifier's character set support.

[Rafael Espíndola via llvm-commits]

On 1 March 2016 at 16:02, Rafael Espíndola <rafael.espindola at gmail.com> wrote:
>>> Just isAllowedChar?
>> Nice, may be isAllowedIDChar?
>
> Yes, that is fine.
>
>>> > Well it based on my previuos experience on Windows where we had lexer
>>> using these routines eating up to 10% of scan time. Probably not so
>>> "generally" as I stated though. I'm not insisting on this particular change and
>>> can remove it.
>>>
>>> Yes, please leave it out. It can be done as an independent patch.
>>>
>>> BTW, how exactly have you benchmarked this?
>>
>> Ran all MC\*.* tests several times and measured time. To be honest I didn’t benchmarked individual contribution of  isDigit change. It mostly benefits from single "in bitvector" check.
>
> That can be somewhat noisy because it consists of many small files. I
> will try to produce a single big .s and upload it somewhere.

https://drive.google.com/file/d/0B7iRtublysV6RmoyOWlhV2I2aXM/view?usp=sharing


That is the .s for the entirety of clang :-)

Cheers,
Rafael