Improve incremental compile speed on webrender_api

[nnethercote]

This was originally filed as a rustc-perf issue, but I'm moving it here because it's really a rustc issue.

The workload in question:

cd webrender/webrender_api && cargo check && touch color.rs && cargo check

In rustc-perf parlance, that is a "CleanIncr" build.

The 5692 lines of rust code in webrender_api expand to 85511 with cargo expand, largely due to serde, and the expansion phase dominates. So it's basically a stress test for proc macro expansion.

#59476 helped a tiny bit. #59507 helped more. The major remaining source of inefficiency is libproc_macro's buffer for communicating between client and server. Buffer::write_all is called more than a million times, almost always for a single byte, mostly due to uleb128 encoding of integers. @eddyb said they would try to improve that code. I estimate this could save another 10%, maybe more.

[nnethercote]

Here is some edited Cachegrind output from the abovementioned workload. I removed various entries that didn't obviously relate to proc_macro::bridge.

--------------------------------------------------------------------------------
Ir                      file:function
--------------------------------------------------------------------------------
1,491,299,765 ( 6.63%)  /build/glibc-B9XfQf/glibc-2.28/string/../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:__memcpy_avx_unaligned_erms
  825,107,571 ( 3.67%)  /home/njn/moz/rust0/src/libproc_macro/bridge/buffer.rs:proc_macro::bridge::buffer::Buffer<T>::extend_from_slice
  364,463,350 ( 1.62%)  /home/njn/moz/rust0/src/libproc_macro/bridge/server.rs:<proc_macro::bridge::server::Dispatcher<proc_macro::bridge::server::MarkedTypes<S>> as proc_macro::bridge::server::DispatcherTrait>::dispatch
  294,650,515 ( 1.31%)  /home/njn/moz/rust0/src/libcore/slice/mod.rs:proc_macro::bridge::buffer::Buffer<T>::extend_from_slice
  249,165,011 ( 1.11%)  /home/njn/moz/rust0/src/libproc_macro/bridge/rpc.rs:<proc_macro::bridge::server::Dispatcher<proc_macro::bridge::server::MarkedTypes<S>> as proc_macro::bridge::server::DispatcherTrait>::dispatch
  222,085,140 ( 0.99%)  /home/njn/moz/rust0/src/libproc_macro/bridge/buffer.rs:<proc_macro::bridge::buffer::Buffer<u8> as std::io::Write>::write_all
  207,506,745 ( 0.92%)  /home/njn/moz/rust0/src/libproc_macro/bridge/rpc.rs:<core::result::Result<T,E> as proc_macro::bridge::rpc::DecodeMut<S>>::decode
  141,654,879 ( 0.63%)  /home/njn/moz/rust0/src/liballoc/collections/btree/search.rs:proc_macro::bridge::client::<impl proc_macro::bridge::rpc::DecodeMut<proc_macro::bridge::client::HandleStore<proc_macro::bridge::server::MarkedTypes<S>>> for proc_macro::bridge::Marked<<S as proc_macro::bridge::server::Types>::TokenStream,proc_macro::bridge::client::TokenStream>>::decode
  138,111,307 ( 0.61%)  /home/njn/moz/rust0/src/libproc_macro/bridge/rpc.rs:<core::result::Result<T,E> as proc_macro::bridge::rpc::Encode<S>>::encode
  125,087,401 ( 0.56%)  /home/njn/moz/rust0/src/libproc_macro/bridge/rpc.rs:<proc_macro::bridge::server::Dispatcher<proc_macro::bridge::server::MarkedTypes<S>> as proc_macro::bridge::server::DispatcherTrait>::dispatch::{{closure}}
  121,934,434 ( 0.54%)  /home/njn/moz/rust0/src/liballoc/collections/btree/search.rs:proc_macro::bridge::client::<impl proc_macro::bridge::rpc::Encode<proc_macro::bridge::client::HandleStore<proc_macro::bridge::server::MarkedTypes<S>>> for proc_macro::bridge::Marked<<S as proc_macro::bridge::server::Types>::TokenStream,proc_macro::bridge::client::TokenStream>>::encode
  117,864,212 ( 0.52%)  /home/njn/moz/rust0/src/libcore/num/mod.rs:proc_macro::bridge::buffer::Buffer<T>::extend_from_slice
  117,647,691 ( 0.52%)  /home/njn/moz/rust0/src/libstd/panicking.rs:<proc_macro::bridge::server::Dispatcher<proc_macro::bridge::server::MarkedTypes<S>> as proc_macro::bridge::server::DispatcherTrait>::dispatch

[nnethercote]

#59820 helps some more.

[Mark-Simulacrum]

Closing. The particular benchmark I think was removed & replaced in ~2022, but the current incr-unchanged compile times are around 1 second for this crate, and have been improving steadily. I don't think there's necessarily more to pay attention to for this particular case.

(src: https://perf.rust-lang.org/index.html?start=2019-04-01&benchmark=webrender-2022&profile=check&scenario=incr-unchanged&stat=wall-time&kind=raw)