Optimize Write::write_fmt for Vec<u8>/Sink

[sinkuu]

A small optimization for formatted writing to Vec<u8>. This doesn't fix the issue, but benchmarks for #10761 (comment) :

 name                default ns/iter  override ns/iter  diff ns/iter   diff % 
 bench_write_macro1  22,253           17,460                  -4,793  -21.54% 
 bench_write_macro2  33,725           29,363                  -4,362  -12.93% 
 bench_write_ref     2,303            2,228                      -75   -3.26% 
 bench_write_value   2,263            2,252                      -11   -0.49%

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @aturon (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

[bluss]

Do you know roughly why this is an improvement?

[alexcrichton]

Hm yeah I'm also a little baffled on how this could result in a speedup. The #[inline] tags on Sink make sense to me, but AFAIK the specializations of write_fmt should be what's generated in the standard case anyway...

[sinkuu]

Looking at IR diff, the significant difference looks like write_fmt beging inlined. But, just putting #[inline] on default write_fmt did't improve performance IIRC (I'll try it again).

https://gist.github.com/sinkuu/9f6c5e34e05c088e84c7330e2b146585

Tried: #[inline} on default write_fmt doesn't improve performance...

 name                old ns/iter  inline ns/iter  diff ns/iter  diff % 
 bench_write_macro1  24,012       24,130                   118   0.49% 
 bench_write_macro2  35,795       36,072                   277   0.77% 
 bench_write_ref     2,374        2,418                     44   1.85% 
 bench_write_value   2,410        2,290                   -120  -4.98% 

[bluss]

It's a very small program (the benchmark runner), so whole program optimization effects could cause this.

[sfackler]

Why do we particularly care about the performance of Sink?

[sinkuu]

I tried to remove unnecessary error handling in write_fmt, and thought that this can be applied to Vec<u8> and Sink.

Since the optimizer seems to be able to do that (and the cause of performance gain was inlining), this PR may not make much sense.