store ObligationCause on the heap

[lcnr]

Stores ObligationCause on the heap using an Rc.

This PR trades off some transient memory allocations to reduce the size of–and thus the number of instructions required to memcpy–a few widely used data structures in trait solving.

[rust-highfive]

r? @ecstatic-morse

(rust_highfive has picked a reviewer for you, use r? to override)

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 91f5ff558068ff7659e7d7c87b62594ffeb9ca7f with merge 7e933e1ada25d4898101cc7e4e2fa77498f948e0...

[jonas-schievink]

@bors try-

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit aef40a48b1750ead607eef1a4d7b19c345ea200a with merge 8ff82fab989f4a07dfd168708da2ffc413addb04...

[bors]

☀️ Try build successful - checks-azure
Build commit: 8ff82fab989f4a07dfd168708da2ffc413addb04 (8ff82fab989f4a07dfd168708da2ffc413addb04)

[rust-timer]

Queued 8ff82fab989f4a07dfd168708da2ffc413addb04 with parent 56daaf6, future comparison URL.

[rust-timer]

Finished benchmarking try commit (8ff82fab989f4a07dfd168708da2ffc413addb04): comparison url.

[lcnr]

This perf is interesting. I think this PR would be worth it as is. I might try and look a bit more into why some benchmarks are negative and what can be done here...

[jonas-schievink]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 61429d6b01a297568decae2ea552e8df6710ff2b with merge ba0ecc9e775e63cb1e782322e6ead5f8b018a463...

[bors]

☀️ Try build successful - checks-azure
Build commit: ba0ecc9e775e63cb1e782322e6ead5f8b018a463 (ba0ecc9e775e63cb1e782322e6ead5f8b018a463)

[rust-timer]

Queued ba0ecc9e775e63cb1e782322e6ead5f8b018a463 with parent 3d5d0f8, future comparison URL.

[rust-timer]

Finished benchmarking try commit (ba0ecc9e775e63cb1e782322e6ead5f8b018a463): comparison url.

[jonas-schievink]

could not find start commit for bound Commit("3d5d0f898c2f3998e50c2180c6202f193c3acdbc")

:(

[mati865]

@jonas-schievink it's still in the queue: https://perf.rust-lang.org/status.html but it has been:

Currently benchmarking: 8ff82fab989f4
37 benchmarks left out of 38 total

for 2 hours so it appears to be stuck.

[Mark-Simulacrum]

Sorry, we were migrating to a new backend for the site (and somewhat the collector), the link(s) should be good to go now.

[lcnr]

Perf is looking fairly good already, I want to look into 1 more thing before this is merged though

[ecstatic-morse]

Thank you!

@bors r+ rollup=never

[bors]

📌 Commit 57fc011ae4ff3c925d61b8b8cf7c252b3dfb6705 has been approved by ecstatic-morse

[ecstatic-morse]

Oh, and if you feel like squashing this, ping me and I'll re-approve.

[lcnr]

@ecstatic-morse done 👍

[ecstatic-morse]

@bors r+

[bors]

📌 Commit ea668d9 has been approved by ecstatic-morse

[nnethercote]

@nnethercote, do you want me to ping you on stuff like this?

I'm happy to be CC'd on things like this, thanks.

[Dylan-DPC-zz]

@bors p=1

[bors]

⌛ Testing commit ea668d9 with merge c8a9c34...

[bors]

☀️ Test successful - checks-azure
Approved by: ecstatic-morse
Pushing c8a9c34 to master...

[nnethercote]

Perf results from landing show a win, as expected. Nice work!

[nnethercote]

This was a clear perf win, but the number of additional allocations done is significant. DHAT shows me that, on numerous benchmarks, around 5% of all allocations done are for ObligationCause::misc. I'm wondering if that could be improved somehow.

[lcnr]

I didn't yet try storing only the ObligationCauseCode on the heap, this would increase the size of Obligation by 16 (still 24 less than the original size), but probably saves a lot of allocations if we don't alloc for ObligationCauseCode::MiscObligation.

I don't have the time to try this myself though

[nnethercote]

I didn't yet try storing only the ObligationCauseCode on the heap, this would increase the size of Obligation by 16 (still 24 less than the original size), but probably saves a lot of allocations if we don't alloc for ObligationCauseCode::MiscObligation.

I've tried this in #73983.