Polymorphization #69749

davidtwco · 2020-03-05T22:33:27Z

This PR implements an analysis to detect when functions could remain polymorphic during code generation.

Fixes #46477

r? @eddyb
cc @rust-lang/wg-mir-opt @nikomatsakis @pnkfelix

Centril

Some minor things...

src/librustc/ty/instance.rs

src/librustc_codegen_llvm/callee.rs

src/librustc_mir/monomorphize/polymorphize.rs

src/librustc_mir/monomorphize/collector.rs

eddyb · 2020-03-06T09:03:12Z

src/librustc/ty/instance.rs

-        if self.substs.has_param_types() {
+        if self.substs.has_param_types() && !tcx.sess.opts.debugging_opts.polymorphize {
            bug!("Instance.ty called for type {:?} with params in substs: {:?}", ty, self.substs);


I forget where/if I left a comment, but the current situation with Instance's types and Instance::ty_env in particular is pretty bad and we should probably require passing ParamEnv around a lot (or maybe start using ParamEnvAnd<Instance>?).

I think it was #67800 (comment).

src/librustc/ty/instance.rs

src/librustc/query/mod.rs

src/librustc/ty/normalize_erasing_regions.rs

src/librustc/query/mod.rs

bjorn3 · 2020-03-06T11:01:54Z

Why does this use ty::Param instead of something like ty::Erased? Is it to reduce the amount of places that you need to change?

eddyb · 2020-03-06T11:11:48Z

Why does this use ty::Param instead of something like ty::Erased? Is it to reduce the amount of places that you need to change?

Using ty::Param should result in more cache reuse, but also it's known to work, since type-checking, miri and other parts of the compiler already extensively rely on it.

Also, if the parameters are truly unused then we might get away without a ParamEnv, but long-term we can rely on ParamEnv to handle parameters used only in certain ways, e.g. #46477 (comment)

src/librustc_codegen_llvm/callee.rs

src/librustc_codegen_llvm/debuginfo/metadata.rs

src/librustc_codegen_llvm/mono_item.rs

src/librustc_codegen_ssa/meth.rs

src/librustc_codegen_utils/symbol_names/legacy.rs

src/librustc_mir/monomorphize/polymorphize.rs

This commit implements the `unused_generic_params` query, an initial version of polymorphization which detects when an item does not use generic parameters and is being needlessly monomorphized as a result. Signed-off-by: David Wood <david@davidtw.co>

This commit normalizes function signatures for instances before substituting, a workaround for polymorphization considering parameters unused when they show up in the signature, but not the body (due to being normalized). Unfortunately, this causes test output to change with the parallel compiler only. Signed-off-by: David Wood <david@davidtw.co>

This commit records the results of `unused_generic_params` in crate metadata, hopefully improving performance. Signed-off-by: David Wood <david@davidtw.co>

This commit replaces the `-Z polymorphize-errors` debugging flag with a `#[rustc_polymorphize_error]` attribute for use on functions. Signed-off-by: David Wood <david@davidtw.co>

This commit introduces a `FiniteBitSet` type which replaces the manual bit manipulation which was being performed in polymorphization. Signed-off-by: David Wood <david@davidtw.co>

davidtwco · 2020-07-20T20:43:29Z

@bors r=eddyb

bors · 2020-07-20T20:43:30Z

📌 Commit 4b99699 has been approved by eddyb

bors · 2020-07-20T20:43:31Z

🌲 The tree is currently closed for pull requests below priority 5, this pull request will be tested once the tree is reopened

bors · 2020-07-21T03:09:10Z

⌛ Testing commit 4b99699 with merge b52522a...

bors · 2020-07-21T06:55:04Z

☀️ Test successful - checks-actions, checks-azure
Approved by: eddyb
Pushing b52522a to master...

rust-highfive · 2020-07-21T06:56:33Z

📣 Toolstate changed by #69749!

Tested on commit b52522a.
Direct link to PR: #69749

💔 miri on windows: test-pass → build-fail (cc @oli-obk @eddyb @RalfJung).
💔 miri on linux: test-pass → build-fail (cc @oli-obk @eddyb @RalfJung).

@oli-obk

Tested on commit rust-lang/rust@b52522a. Direct link to PR: <rust-lang/rust#69749> 💔 miri on windows: test-pass → build-fail (cc @oli-obk @eddyb @RalfJung). 💔 miri on linux: test-pass → build-fail (cc @oli-obk @eddyb @RalfJung).

@eddyb

rustup Following rust-lang/rust#69749 I added some `ty::ParamEnv::reveal_all()` even though @eddyb advised me in the past to avoid those.

nnethercote · 2020-07-27T23:24:36Z

This was a perf loss on landing, as expected.

rust-highfive assigned eddyb Mar 5, 2020

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Mar 5, 2020

This comment has been minimized.

Sign in to view

Centril reviewed Mar 6, 2020

View reviewed changes