Shrink Expr_::ExprInlineAsm.

[nnethercote]

On 64-bit this reduces the size of Expr_ from 144 to 64 bytes, and
reduces the size of Expr from 176 to 96 bytes.

For the workload in #36799 this reduces the RSS for the "lowering ast -> hir" phase and all subsequent phases by 50 MiB, which reduces the peak RSS for that workload by about 1%. Not huge, but it's a very easy improvement.

r? @eddyb

[eddyb]

cc @jseyfried Doing the same for the AST might speed up expansion.

@bors r+

[bors]

📌 Commit a920e35 has been approved by eddyb

[bors]

⌛ Testing commit a920e35 with merge f5a702d...

[bors]

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-64-cargotest, auto-linux-64-cross-freebsd, auto-linux-64-debug-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-opt-rustbuild, auto-linux-64-x-android-t, auto-linux-cross-opt, auto-linux-musl-64-opt, auto-mac-32-opt, auto-mac-64-opt, auto-mac-64-opt-rustbuild, auto-mac-cross-ios-opt, auto-win-gnu-32-opt, auto-win-gnu-32-opt-rustbuild, auto-win-gnu-64-opt, auto-win-msvc-32-opt, auto-win-msvc-64-cargotest, auto-win-msvc-64-opt, auto-win-msvc-64-opt-rustbuild
Approved by: eddyb
Pushing f5a702d to master...

[nnethercote]

For the workload in #36799 this reduces the RSS for the "lowering ast -> hir" phase and all subsequent phases by 50 MiB, which reduces the peak RSS for that workload by about 1%. Not huge, but it's a very easy improvement.

I did some more measurements because the above improvement was smaller than expected. Massif showed that this change does make a decent difference. Here are measurements for a cut down version of ostn15_phf that peaks at ~1 GiB instead of ~4.5 GiB. Before:

98.81% (1,053,399,475B) (heap allocation functions) malloc/new/new[], --alloc-fns, etc.
->23.11% (246,413,376B) 0x7E07CE6: rustc::hir::lowering::LoweringContext::lower_expr (heap.rs:59)
| ->09.91% (105,601,584B) 0x7DC80AD: <collections::vec::Vec<T> as core::iter::traits::FromIterator<T>>::from_iter (lowering.rs:1036)

After:

98.69% (949,710,821B) (heap allocation functions) malloc/new/new[], --alloc-fns, etc.
->13.97% (134,407,296B) 0x7E09522: rustc::hir::lowering::LoweringContext::lower_expr (heap.rs:59)
| ->05.99% (57,600,864B) 0x7DC9B9E: <collections::vec::Vec<T> as core::iter::traits::FromIterator<T>>::from_iter (lowering.rs:1036)

The Expr allocations are reduced in size by almost 50%, as expected. Good.

I then did some RSS measurements of stage2 compilers via /usr/bin/time. They jump around a bit, but if I take the lowest of 3 runs, the before vs. after improvement is 1108 MB -> 1004 MB, which is a 9% drop. Good.

In conclusion, I'm not sure what went wrong with my initial measurements, but I'm confident now that this reduces memory usage for ostn15_phf by about 9%.

[brson]

Nice wins.