Move query code outside macros and store query jobs separately from query results

[Zoxc]

Based on #50067

r? @michaelwoerister

[michaelwoerister]

Thanks, @Zoxc!

Do you have any measurements how splitting the map affects memory usage and performance? In the cache miss case, it leads to a few more hashtable lookups than the current version, I think.

[Zoxc]

It uses 450MB when compiling syntex_syntax vs. 456MB before. Compile times seems to be slightly reduced, although that may just be due to noise.

[TimNN]

Ping from triage @michaelwoerister ! This PR needs your review.

[Zoxc]

@bors try

[bors]

⌛ Trying commit 87956b5a5eaee00b89297551c1dd75e1ff632650 with merge 34c50be0a1c561e15f7d02e06c7e47b48a40e444...

[bors]

☀️ Test successful - status-travis
State: approved= try=True

[Zoxc]

@Mark-Simulacrum I'd like a perf run here

[Mark-Simulacrum]

Perf queued.

[Zoxc]

I did some microbenchmarks here too:
Before:

  time: 0.827; rss: 36MB        hot query (usize)
  time: 0.942; rss: 170MB       cold query (usize)
  time: 0.911; rss: 170MB       hot query (dummy DefId)
  time: 0.972; rss: 304MB       cold query (dummy DefId)
  time: 0.960; rss: 304MB       hot query (DefId)
  time: 0.953; rss: 438MB       cold query (DefId)

After:

  time: 0.367; rss: 36MB        hot query (usize)
  time: 0.717; rss: 136MB       cold query (usize)
  time: 0.498; rss: 136MB       hot query (DefId)
  time: 0.795; rss: 237MB       cold query (DefId)
  time: 0.379; rss: 237MB       hot query (dummy DefId)
  time: 0.706; rss: 338MB       cold query (dummy DefId)

Before (with incremental compilation):

  time: 0.958; rss: 40MB        hot query (usize)
  time: 3.165; rss: 638MB       cold query (usize)
  time: 1.080; rss: 638MB       hot query (dummy DefId)
  time: 3.286; rss: 1237MB      cold query (dummy DefId)
  time: 1.088; rss: 1237MB      hot query (DefId)

After (with incremental compilation):

  time: 0.511; rss: 40MB        hot query (usize)
  time: 2.584; rss: 605MB       cold query (usize)
  time: 0.680; rss: 605MB       hot query (DefId)
  time: 0.478; rss: 605MB       hot query (dummy DefId)
  time: 2.796; rss: 1170MB      cold query (dummy DefId)

[michaelwoerister]

Looks very promising, @Zoxc!

[Zoxc]

Looks green.
http://perf.rust-lang.org/compare.html?start=25749ad66de71ffb46e6b04fb0cccf1ae4fc5ed5&end=34c50be0a1c561e15f7d02e06c7e47b48a40e444&stat=instructions%3Au

I think I'm going to close #50067 in favor of this.

[michaelwoerister]

Thanks, @Zoxc! The first two commits look great. The third one I'm a little ambiguous about because of the additional table lookups. But since performance seems to improve overall, I don't have any real objections. All in all this PR cleans up the core query code considerably! No we just have to merge dep_node.rs into plumbing.rs and all will be great in query land ;)

r=me with the comments from #50067 addressed.

[Zoxc]

@bors r=michaelwoerister

[bors]

📌 Commit f678bf0 has been approved by michaelwoerister

[michaelwoerister]

🎉

[bors]

⌛ Testing commit f678bf0 with merge 3c43aa5...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: michaelwoerister
Pushing 3c43aa5 to master...