Refactor call & function handling in trans, enable MIR bootstrap. by eddyb · Pull Request #32080 · rust-lang/rust

eddyb · 2016-03-06T15:10:23Z

Non-Rust and Rust ABIs were combined into a common codepath, which means:

The ugly __rust_abi "clown shoes" shim for C->Rust FFI is gone, fixesremove the need for foreign ABI wrappers #10116.
Methods, including virtual ones support non-Rust ABIs, closesFeature-gate non-Rust ABI methods #30235.
Non-Rust ABIs also pass fat pointers in two arguments; the result should be identical.
Zero-sized types are never passed as arguments; again, behavior shouldn't change.

Additionally, MIR support for calling intrinsics (through old trans) was implemented.
Alongside assorted fixes, it enabled MIR to launch 🚀 and do a complete bootstrap.
To try it yourself, ./configure --enable-orbitormake RUSTFLAGS="-Z orbit".

alexcrichton · 2016-03-06T18:30:48Z

r? @nikomatsakis

(just making sure an assignment is given, feel free to delegate elsewhere)

bors · 2016-03-08T09:33:11Z

☔ The latest upstream changes (presumably #31606) made this pull request unmergeable. Please resolve the merge conflicts.

eddyb · 2016-03-09T12:16:23Z

Status: I can build snap-stage3 with MIR forced on for everything.
These are the common run-pass failures for stage1 and stage2:
(that means they are likely MIR trans bugs and not due to miscompiled rustc)

failures:
[run-pass] run-pass/allocator-override.rs
[run-pass] run-pass/asm-in-out-operand.rs
[run-pass] run-pass/asm-indirect-memory.rs
[run-pass] run-pass/asm-out-assign.rs
[run-pass] run-pass/associated-const-type-parameters.rs
[run-pass] run-pass/backtrace-debuginfo.rs
[run-pass] run-pass/const-str-ptr.rs
[run-pass] run-pass/ignore-all-the-things.rs
[run-pass] run-pass/int-abs-overflow.rs
[run-pass] run-pass/issue-14936.rs
[run-pass] run-pass/issue-15080.rs
[run-pass] run-pass/issue-15104.rs
[run-pass] run-pass/issue-23338-ensure-param-drop-order.rs
[run-pass] run-pass/issue-7784.rs
[run-pass] run-pass/issue-8460.rs
[run-pass] run-pass/issue-8860.rs
[run-pass] run-pass/match-vec-alternatives.rs
[run-pass] run-pass/mir_raw_fat_ptr.rs
[run-pass] run-pass/num-wrapping.rs
[run-pass] run-pass/out-of-stack.rs
[run-pass] run-pass/simd-intrinsic-generic-cast.rs
[run-pass] run-pass/simd-intrinsic-generic-elements.rs
[run-pass] run-pass/super-fast-paren-parsing.rs
[run-pass] run-pass/vec-matching-fixed.rs
[run-pass] run-pass/vec-matching-fold.rs
[run-pass] run-pass/vec-matching-legal-tail-element-borrow.rs
[run-pass] run-pass/vec-matching.rs
[run-pass] run-pass/vec-tail-matching.rs
[run-pass] run-pass/zero-size-type-destructors.rs

The only run-pass test that only fails in stage2 is run-pass/super-fast-paren-parsing.rs, which ends up in a stack overflow in rustc (@arielb1 was talking about needing llvm lifetime intrinsics to prevent this, I think).

@rust-lang/tools If I provide a flag to enable MIR trans for everything, and an attribute to mark functions which can't be yet properly translated via MIR, would it be possible to have buildbot ensure MIR doesn't regress?

nikomatsakis · 2016-03-09T14:24:01Z

‼️ 😲

alexcrichton · 2016-03-09T17:14:27Z

@eddyb I'd be fine at least adding a MIR bot! If it's a ./configure switch just lemme know and I can set up something like auto-linux-64-opt-mir or something like that.

Do you think MIR will want platform coverage as well? A Linux/OSX/Windows builder, or just one with it enabled?

eddyb · 2016-03-09T20:25:40Z

@alexcrichton Before the next snapshot, make RUSTFLAGS_STAGE1="-Z orbit" RUSTFLAGS_STAGE2=-"Z orbit", after that just make RUSTFLAGS="-Z orbit".

Sadly, it doesn't look like tests inherit RUSTFLAGS_STAGEn, just RUSTFLAGS, so we might need a snapshot anyway.

We'll want at least a non-MSVC and a MSVC builder, because of the differences in how unwinding works there.

alexcrichton · 2016-03-09T20:27:39Z

Ok, sounds like a linux + MSVC builder, and can you add this as a ./configure option? That'll be easiest to add to our build matrix.

eddyb · 2016-03-09T20:46:53Z

@alexcrichton Does that last commit look good?

alexcrichton · 2016-03-09T22:11:41Z

Looks good to me!

eddyb · 2016-03-10T00:27:07Z

Down to 9 run-pass MIR failures:

 [run-pass] run-pass/backtrace-debuginfo.rs
[run-pass] run-pass/const-str-ptr.rs
[run-pass] run-pass/int-abs-overflow.rs
[run-pass] run-pass/issue-23338-ensure-param-drop-order.rs
[run-pass] run-pass/issue-8460.rs
[run-pass] run-pass/issue-8860.rs
[run-pass] run-pass/mir_raw_fat_ptr.rs
[run-pass] run-pass/simd-intrinsic-generic-elements.rs
[run-pass] run-pass/zero-size-type-destructors.rs

I'll just start tagging them with #[rustc_no_mir], this PR is getting kind of long...

bors · 2016-03-18T12:29:36Z

📌 Commit e177207 has been approved by nikomatsakis

bors · 2016-03-18T12:29:39Z

⌛ Testing commit e177207 with merge 3c572b6...

bors · 2016-03-18T13:49:08Z

💔 Test failed - auto-win-gnu-64-opt

…ssed.

eddyb · 2016-03-18T13:54:56Z

@bors r=nikomatsakis force

bors · 2016-03-18T13:54:56Z

📌 Commit b12dcde has been approved by nikomatsakis

bors · 2016-03-18T13:54:59Z

⌛ Testing commit b12dcde with merge 235d774...

Refactor call & function handling in trans, enable MIR bootstrap. Non-Rust and Rust ABIs were combined into a common codepath, which means: * The ugly `__rust_abi` "clown shoes" shim for C->Rust FFI is gone, fixes#10116. * Methods, *including virtual ones* support non-Rust ABIs, closes#30235. * Non-Rust ABIs also pass fat pointers in two arguments; the result should be identical. * Zero-sized types are never passed as arguments; again, behavior shouldn't change. Additionally, MIR support for calling intrinsics (through old trans) was implemented. Alongside assorted fixes, it enabled MIR to launch 🚀 and do a *complete* bootstrap. To try it yourself, `./configure --enable-orbit` *or* `make RUSTFLAGS="-Z orbit"`.

bors · 2016-03-18T16:06:09Z

jonas-schievink · 2016-03-18T16:29:47Z

It's doing strictly more work, AFAICT, because we emit more IR (not sure exactly why yet).

Pretty impressive that translation still ends up being faster!

@nikomatsakis

Update snapshots to 2016-03-18 (235d774). The new snapshots notably include @nikomatsakis' #32164 fix for the regionck error spew. Also, `./configure --enable-orbit` *and/or* `make RUSTFLAGS="-Z orbit"` should work now (#32080).

rBreak Critical Edges and other MIR work This PR is built on top of #32080. This adds the basic depth-first traversals for MIR, preorder, postorder and reverse postorder. The MIR blocks are now translated using reverse postorder. There is also a transform for breaking critical edges, which includes the edges from `invoke`d calls (`Drop` and `Call`), to account for the fact that we can't add code after an `invoke`. It also stops generating the intermediate block (since the transform essentially does it if necessary already). The kinds of cases this deals with are difficult to produce, so the test is the one I managed to get. However, it seems to bootstrap with `-Z orbit`, which it didn't before my changes.

eddyb force-pushed the transcendent branch from 2fbb50c to 90130dcCompare March 6, 2016 15:35

alexcrichton assigned nikomatsakis Mar 6, 2016

eddyb force-pushed the transcendent branch from 90130dc to ee6c793Compare March 6, 2016 20:04

This was referenced Mar 6, 2016
Distinguish fn item types to allow reification from nothing to fn pointers. #31710
Merged
[MIR] Reconsider invoked function return value strategy #32105
Closed

eddyb force-pushed the transcendent branch from b7e69e9 to 40f3bc9Compare March 9, 2016 12:22

eddyb force-pushed the transcendent branch 2 times, most recently from 934c483 to 5024b6dCompare March 9, 2016 14:49

eddyb force-pushed the transcendent branch from 5024b6d to c8ba80bCompare March 9, 2016 20:24

eddyb changed the title ~~[WIP] Refactor how calls and functions are handled across trans.~~[WIP] Refactor call & function handling in trans, enable MIR bootstrap.Mar 9, 2016

eddyb force-pushed the transcendent branch from c8ba80b to 23b378dCompare March 9, 2016 20:46

eddyb force-pushed the transcendent branch from 7d5f75a to afcafe6Compare March 9, 2016 22:16

eddyb mentioned this pull request Mar 10, 2016
Mir: casting &str to *const str will cast twice #30612
Closed

eddyb changed the title ~~[WIP] Refactor call & function handling in trans, enable MIR bootstrap.~~Refactor call & function handling in trans, enable MIR bootstrap.Mar 10, 2016

eddyb force-pushed the transcendent branch 6 times, most recently from 76a4fa2 to 0c413a8Compare March 10, 2016 14:56

tests: Update run-make/issue-25581 to reflect how fat pointers are pa…
b12dcde
…ssed.

bors merged commit b12dcde into rust-lang:masterMar 18, 2016

eddyb deleted the transcendent branch March 18, 2016 16:06

This was referenced Mar 18, 2016
Revamp symbol names for impls (and make them deterministic, etc) #32293
Merged
mk: Add configure option for disabling codegen tests #32168
Merged
configure: Support --disable-option-checking #31169
Merged

This was referenced Mar 19, 2016
Update snapshots to 2016-03-18 (235d774). #32345
Merged
Add (temporary) MIR builders. rust-lang-deprecated/rust-buildbot#82
Closed

nasa42 mentioned this pull request Mar 21, 2016
notable changes and CotW for TWIR 123 rust-lang/this-week-in-rust#187
Merged

eddyb mentioned this pull request Mar 24, 2016
Wrong codegen for generic foreign functions. #32364
Closed

JohanLorenzo mentioned this pull request Apr 5, 2016
Can't compile the nix library since da66431 #32740
Closed

eddyb mentioned this pull request Jun 7, 2016
LLVM error while cross-compiling Servo for ARM #32049
Closed

pnkfelix mentioned this pull request Jun 8, 2016
ICE: OutputTypeParameterMismatch in failing to resolve associated type as tuple. #33364
Closed

arielb1 mentioned this pull request Oct 21, 2016
Compiler hangs after 'write metadata stage' #37311
Closed

nbdd0121 mentioned this pull request Jul 1, 2022
Optimize layout of function arguments in the Rust ABI - take 2 #97559
Closed

RalfJung mentioned this pull request Jan 8, 2025
fix handling of ZST in win64 ABI on windows-msvc targets #135204
Merged

Uh oh!

Uh oh!

Conversation

eddyb commented Mar 6, 2016

Uh oh!

alexcrichton commented Mar 6, 2016

Uh oh!

bors commented Mar 8, 2016

Uh oh!

eddyb commented Mar 9, 2016

Uh oh!

nikomatsakis commented Mar 9, 2016

Uh oh!

alexcrichton commented Mar 9, 2016

Uh oh!

eddyb commented Mar 9, 2016

Uh oh!

alexcrichton commented Mar 9, 2016

Uh oh!

eddyb commented Mar 9, 2016

Uh oh!

alexcrichton commented Mar 9, 2016

Uh oh!

eddyb commented Mar 10, 2016

Uh oh!

bors commented Mar 18, 2016

Uh oh!

bors commented Mar 18, 2016

Uh oh!

bors commented Mar 18, 2016

Uh oh!

eddyb commented Mar 18, 2016

Uh oh!

bors commented Mar 18, 2016

Uh oh!

bors commented Mar 18, 2016

Uh oh!

bors commented Mar 18, 2016

Uh oh!

jonas-schievink commented Mar 18, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants