Mark other variants as uninitialized after switch on discriminant

[ecstatic-morse]

During drop elaboration, which builds the drop ladder that handles destruction during stack unwinding, we attempt to remove MIR Drop terminators that will never be reached in practice. This reduces the number of basic blocks that are passed to LLVM, which should improve performance. In #66753, a user pointed out that unreachable Drop terminators are common in functions like Option::unwrap, which move out of an enum. While discussing possible remedies for that issue, @eddyb suggested moving const-checking after drop elaboration. This would allow the former, which looks for Drop terminators and replicates a small amount of drop elaboration to determine whether a dropped local has been moved out, leverage the work done by the latter.

However, it turns out that drop elaboration is not as precise as it could be when it comes to eliminating useless drop terminators. For example, let's look at the code for unwrap_or.

fn unwrap_or<T>(opt: Option<T>, default: T) -> T {
    match opt {
        Some(inner) => inner,
        None => default,
    }
}

opt never needs to be dropped, since it is either moved out of (if it is Some) or has no drop glue (if it is None), and default only needs to be dropped if opt is Some. This is not reflected in the MIR we currently pass to codegen.

@eddyb also suggested the solution to this problem. When we switch on an enum discriminant, we should be marking all fields in other variants as definitely uninitialized. I implemented this on top of alongside a small optimization (split out into #68943) that suppresses drop terminators for enum variants with no fields (e.g. Option::None). This is the resulting MIR for unwrap_or.

In concert with #68943, this change speeds up many optimized and debug builds. We need to carefully investigate whether I have introduced any miscompilations before merging this. Code that never drops anything would be very fast indeed until memory is exhausted.

[rust-highfive]

The job x86_64-gnu-llvm-7 of your PR failed (pretty log, raw log). Through arcane magic we have determined that the following fragments from the build log may contain information about the problem.

Click to expand the log.

2020-01-25T00:29:26.5085565Z ========================== Starting Command Output ===========================
2020-01-25T00:29:26.5090369Z [command]/bin/bash --noprofile --norc /home/vsts/work/_temp/4cf1a6f3-5c62-48f1-912d-d20290d394d7.sh
2020-01-25T00:29:26.5090656Z 
2020-01-25T00:29:26.5093479Z ##[section]Finishing: Disable git automatic line ending conversion
2020-01-25T00:29:26.5098584Z ##[section]Starting: Checkout rust-lang/rust@refs/pull/68528/merge to s
2020-01-25T00:29:26.5100065Z Task         : Get sources
2020-01-25T00:29:26.5100093Z Description  : Get sources from a repository. Supports Git, TfsVC, and SVN repositories.
2020-01-25T00:29:26.5100118Z Version      : 1.0.0
2020-01-25T00:29:26.5100143Z Author       : Microsoft
---
2020-01-25T00:29:27.5029435Z ##[command]git remote add origin https://github.com/rust-lang/rust
2020-01-25T00:29:27.5040134Z ##[command]git config gc.auto 0
2020-01-25T00:29:27.5043486Z ##[command]git config --get-all http.https://github.com/rust-lang/rust.extraheader
2020-01-25T00:29:27.5046159Z ##[command]git config --get-all http.proxy
2020-01-25T00:29:27.5052754Z ##[command]git -c http.extraheader="AUTHORIZATION: basic ***" fetch --force --tags --prune --progress --no-recurse-submodules --depth=2 origin +refs/heads/*:refs/remotes/origin/* +refs/pull/68528/merge:refs/remotes/pull/68528/merge
---
2020-01-25T00:52:49.1214028Z    Compiling rustc_index v0.0.0 (/checkout/src/librustc_index)
2020-01-25T00:52:51.2179474Z error: failed to run custom build command for `backtrace-sys v0.1.32`
2020-01-25T00:52:51.2179565Z 
2020-01-25T00:52:51.2179607Z Caused by:
2020-01-25T00:52:51.2180394Z   process didn't exit successfully: `/checkout/obj/build/x86_64-unknown-linux-gnu/stage1-rustc/release/build/backtrace-sys-227a27b0f872ec81/build-script-build` (signal: 11, SIGSEGV: invalid memory reference)
2020-01-25T00:52:51.2180827Z --- stdout
2020-01-25T00:52:51.2181082Z cargo:rustc-cfg=rbt
2020-01-25T00:52:51.2187011Z TARGET = Some("x86_64-unknown-linux-gnu")
2020-01-25T00:52:51.2187085Z OPT_LEVEL = Some("2")
2020-01-25T00:52:51.2187294Z HOST = Some("x86_64-unknown-linux-gnu")
2020-01-25T00:52:51.2187512Z CC_x86_64-unknown-linux-gnu = Some("sccache cc")
2020-01-25T00:52:51.2187751Z warning: build failed, waiting for other jobs to finish...
2020-01-25T00:52:51.7306709Z error: build failed
2020-01-25T00:52:51.7324740Z command did not execute successfully: "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0/bin/cargo" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0/bin/cargo" "build" "-Zconfig-profile" "--target" "x86_64-unknown-linux-gnu" "-Zbinary-dep-depinfo" "-j" "2" "--release" "--locked" "--color" "always" "--features" " llvm" "--manifest-path" "/checkout/src/rustc/Cargo.toml" "--message-format" "json-render-diagnostics"
2020-01-25T00:52:51.7325391Z expected success, got: exit code: 101
2020-01-25T00:52:51.7325391Z expected success, got: exit code: 101
2020-01-25T00:52:51.7336184Z failed to run: /checkout/obj/build/bootstrap/debug/bootstrap test
2020-01-25T00:52:51.7337030Z Build completed unsuccessfully in 0:18:22
2020-01-25T00:52:51.7381164Z == clock drift check ==
2020-01-25T00:52:51.7395424Z   local time: Sat Jan 25 00:52:51 UTC 2020
2020-01-25T00:52:52.0071805Z   network time: Sat, 25 Jan 2020 00:52:52 GMT
2020-01-25T00:52:52.0080702Z == end clock drift check ==
2020-01-25T00:52:52.7690170Z 
2020-01-25T00:52:52.7771012Z ##[error]Bash exited with code '1'.
2020-01-25T00:52:52.7789148Z ##[section]Finishing: Run build
2020-01-25T00:52:52.7800641Z ##[section]Starting: Checkout rust-lang/rust@refs/pull/68528/merge to s
2020-01-25T00:52:52.7802137Z Task         : Get sources
2020-01-25T00:52:52.7802172Z Description  : Get sources from a repository. Supports Git, TfsVC, and SVN repositories.
2020-01-25T00:52:52.7802207Z Version      : 1.0.0
2020-01-25T00:52:52.7802254Z Author       : Microsoft
2020-01-25T00:52:52.7802254Z Author       : Microsoft
2020-01-25T00:52:52.7802288Z Help         : [More Information](https://go.microsoft.com/fwlink/?LinkId=798199)
2020-01-25T00:52:52.7802325Z ==============================================================================
2020-01-25T00:52:53.1143397Z Cleaning any cached credential from repository: rust-lang/rust (GitHub)
2020-01-25T00:52:53.1182947Z ##[section]Finishing: Checkout rust-lang/rust@refs/pull/68528/merge to s
2020-01-25T00:52:53.1268109Z Cleaning up task key
2020-01-25T00:52:53.1268730Z Start cleaning up orphan processes.
2020-01-25T00:52:53.1353669Z Terminate orphan process: pid (5563) (python)
2020-01-25T00:52:53.1604548Z ##[section]Finishing: Finalize Job

I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact @TimNN. (Feature Requests)

[ecstatic-morse]

Just to see how slow this makes dataflow:

@bors try
@rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 679b4238e793c197eacfe43b7fbc4781e948c36b with merge 3828cd570d1515093538246582353d44c84220db...

[bors]

☀️ Try build successful - checks-azure
Build commit: 3828cd570d1515093538246582353d44c84220db (3828cd570d1515093538246582353d44c84220db)

[rust-timer]

Queued 3828cd570d1515093538246582353d44c84220db with parent 8ad83af, future comparison URL.

[rust-timer]

Finished benchmarking try commit 3828cd570d1515093538246582353d44c84220db, comparison URL.

[ecstatic-morse]

@bors try
@rust-timer queue

[rust-timer]

Awaiting bors try build completion

[bors]

⌛ Trying commit 64b1424dfbf7edd2d3b0e2190fbd85bb93753d8c with merge 0077a7aa11ebc2462851676f9f464d5221b17d6a...

[bors]

☀️ Try build successful - checks-azure
Build commit: 0077a7aa11ebc2462851676f9f464d5221b17d6a (0077a7aa11ebc2462851676f9f464d5221b17d6a)

[rust-timer]

Queued 0077a7aa11ebc2462851676f9f464d5221b17d6a with parent d55f3e9, future comparison URL.

[rust-timer]

Finished benchmarking try commit 0077a7aa11ebc2462851676f9f464d5221b17d6a, comparison URL.

[bjorn3]

This reduces the amount of time spent in LLVM. For incremental builds is also reduces the amount of time spent in incr_comp_load_dep_graph by up to 20%. For clean incremental there are a few regressions of ~2%, but those regressions are less than 10ms, so they are likely. The self profiling data even says for some of the regressions that they were actually improvements.

[ecstatic-morse]

@bjorn3 I added a summary of the changes to the OP, along with an explanation of the perf results for builds that don't involve codegen. They are indeed made slightly slower by this PR. A 10% improvement for syn-opt seems almost too good to be true. In fact, it seems so good that I was hesitant to call attention to these results lest this PR later be found to be unsound. If that improvement is legitimate, I would be overjoyed.

[ecstatic-morse]

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

Although there's no point in reviewing until #68241 lands.

[eddyb]

Since I suggested this approach, I'd rather someone else review:
r? @pnkfelix or @matthewjasper cc @arielb1

[rust-highfive]

📣 Toolstate changed by #68528!

Tested on commit 49c68bd.
Direct link to PR: #68528

💔 book on windows: test-pass → test-fail (cc @carols10cents @steveklabnik).