Small improvement for String.length

[abelbraaksma]

The PR removes some overhead for String.length, see #9390 (comment).

The disassembly will change from this:

testrax,rax ; rax & rax, in other words: null-check, sets flagsje short M00_L01 ; if zero (null) jump back to loopcmp dword ptr [rax+8],0 ; compare string-length to 0jbe short M00_L01 ; if equal, jump back to loopmovecx,[rax+8] ; if not, get string length fallthrough to loop

to this:

testrax,rax ; rax & rax, for null-check, sets flagsje short M00_L01 ; if zero (null) jump back to loopmovecx,[rax+8] ; get the string-length from the string and fallthrough to loop

Which, in timings, gives roughly a 33% improvement, but since this is hard to test in isolation, it may be more because more chances exist now for optimizing & inlining this by the JIT (which, in this isolated case, indeed happens).

More info and timings in the connected issue comment above.

[abelbraaksma]

Hmm, I've contributed before, though it's a while ago that I made PR's. From time to time you have to sign again?

[cartermp]

The CLA is a consequence of moving to dotnet/fsharp I believe?

[abelbraaksma]

Possible, anyway, signing took 3 seconds, so it's not a big deal :)

[KevinRansom]

@abelbraaksma It seems as though you combined two changes into one PR and proposed it twice.

/cc @vzarytovskii

[abelbraaksma]

@KevinRansom, yeah, that's why I shouldn't do this long after midnight ;)

I meant to split them (@cartermp and @forkiasked for a PR per function), and thought I only locally had done the accidental merge.

I used VS's github integration, and couldn't switch branches on the PR page, so I couldn't create a new PR on the 2nd branch, and after some frustration clearly did something else wrong ;)

Will fix.

[cartermp]

Thanks!

[KevinRansom]

@abelbraaksma thanks for taking care of this.

Kevin