always back `ty::Const` by an `Allocation` #58486

oli-obk · 2019-02-15T12:27:47Z

This reduces the number of conversion between Allocation and Value by caching both.

…location

src/librustc/mir/interpret/value.rs

RalfJung · 2019-02-15T14:12:22Z

src/librustc/ty/sty.rs

@@ -2065,7 +2066,7 @@ pub enum LazyConst<'tcx> {
 }

 #[cfg(target_arch = "x86_64")]
-static_assert!(LAZY_CONST_SIZE: ::std::mem::size_of::<LazyConst<'static>>() == 56);
+static_assert!(LAZY_CONST_SIZE: ::std::mem::size_of::<LazyConst<'static>>() == 80);


And this isn't a problem? Let's do a perf run.

src/librustc/ty/sty.rs

src/librustc_mir/const_eval.rs

oli-obk · 2019-02-15T15:15:05Z

@bors try

bors · 2019-02-15T15:15:17Z

⌛ Trying commit b0857a7 with merge dc96b91...

@RalfJung

always back `ty::Const` by an `Allocation` This reduces the number of conversion between `Allocation` and `Value` by caching both. r? @RalfJung

src/librustc/ich/impls_ty.rs

bors · 2019-02-15T17:35:03Z

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

oli-obk · 2019-02-15T18:12:24Z

@rust-timer build dc96b91

rust-timer · 2019-02-15T18:12:25Z

Success: Queued dc96b91 with parent f058741, comparison URL.

rust-timer · 2019-02-15T20:46:48Z

Finished benchmarking try commit dc96b91

RalfJung · 2019-02-15T20:50:44Z

max-rss regressed noticeably, mostly for tuple-stress.

RalfJung · 2019-02-16T09:37:28Z

src/librustc_mir/interpret/memory.rs

@@ -647,15 +646,15 @@ where
            // does not permit code that would break this!
            if self.alloc_map.contains_key(&alloc) {
                // Not yet interned, so proceed recursively
-                self.intern_static(alloc, mutability)?;
+                let _alloc = self.intern_static(alloc, mutability)?;


Why bind this to an unused variable?

RalfJung · 2019-02-16T09:38:40Z

src/librustc_mir/const_eval.rs

+        let field = ecx.mplace_field(down, field.index() as u64)?;
+        let val = match field.layout.abi {
+            layout::Abi::Scalar(..) => {
+                let scalar = ecx.try_read_immediate_from_mplace(field)?.unwrap().to_scalar()?;


Couldn't you use try_read_immediate(field.into())? Then we could avoid exposing that internal function try_read_immediate_from_mplace.

src/librustc_mir/interpret/step.rs

RalfJung · 2019-02-16T09:45:03Z

src/librustc_mir/const_eval.rs

+        };
+        let ptr = Pointer::from(constant.alloc_id);
+        let alloc = constant.alloc;
+        Ok(ty::Const { val, ty: mplace.layout.ty, alloc: Some((alloc, ptr))})


The code above here is basically mplace_to_const, right? And we have the same in const_field, don't we? Couldn't that become a helper function?

RalfJung · 2019-02-16T14:44:26Z

src/librustc_mir/const_eval.rs

    })();
    result.map_err(|error| {
        let err = error_to_const_error(&ecx, error);
+        // FIXME(oli-obk): I believe this is unreachable and we can just ICE here. Since a constant
+        // is checked for validity before being in a place that could pass it to `const_field`,
+        // we can't possibly have errors. All fields have already been checked.


Could/should this FIXME be ported to the other PR?

I'm trying this out locally in a different branch, so there's no reason to add it to the source tree

@RalfJung

Const to op simplification r? @RalfJung alternative to rust-lang#58486

@RalfJung

Const to op simplification r? @RalfJung alternative to rust-lang#58486

@RalfJung

Const to op simplification r? @RalfJung alternative to rust-lang#58486

@RalfJung

Const to op simplification r? @RalfJung alternative to rust-lang#58486

bors · 2019-02-24T09:53:26Z

☔ The latest upstream changes (presumably #58691) made this pull request unmergeable. Please resolve the merge conflicts.

RalfJung · 2019-02-24T16:51:03Z

I believe this got superseded by #58511 and can be closed?

oli-obk added 6 commits February 14, 2019 17:28

Setup ty::Const to allow ConstValue::Scalar to be backed by an al…

7c966de

…location

Fiddle the corresponding allocation through all RawConst uses

a5a5ef1

Make const eval not use the hacky op_to_const function

f5fb34c

Fix equality check for constants

db168bc

Promoteds always have a backing allocation

9a63fc8

Fixup: statics are always backed by an allocation

78c36b9

rust-highfive assigned RalfJung Feb 15, 2019

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Feb 15, 2019

oli-obk changed the title ~~always back ty::Const ty an Allocation~~ always back ty::Const by an Allocation Feb 15, 2019

This comment has been minimized.

Sign in to view

RalfJung reviewed Feb 15, 2019

View reviewed changes

src/librustc/mir/interpret/value.rs Show resolved Hide resolved

RalfJung reviewed Feb 15, 2019

View reviewed changes

src/librustc/ty/sty.rs Show resolved Hide resolved

Update comment and make it truer by actually doing what it says

785302d

RalfJung reviewed Feb 15, 2019

View reviewed changes

src/librustc_mir/const_eval.rs Outdated Show resolved Hide resolved

RalfJung reviewed Feb 15, 2019

View reviewed changes

src/librustc_mir/const_eval.rs Outdated Show resolved Hide resolved

oli-obk added 5 commits February 15, 2019 15:27

Explain the val caching in ty::Const

8b0dd6d

Remove a now-dead function argument

6906379

Mark a function private that is unused anywhere else

1b93eda

Fold an if on a match returning a bool directly into the match

b3731ee

Synchronize StableHash with the updated PartialEq impl

b0857a7

bors added a commit that referenced this pull request Feb 15, 2019

Auto merge of #58486 - oli-obk:import_miri_from_future, r=<try>

dc96b91

always back `ty::Const` by an `Allocation` This reduces the number of conversion between `Allocation` and `Value` by caching both. r? @RalfJung

RalfJung reviewed Feb 15, 2019

View reviewed changes

src/librustc/ich/impls_ty.rs Show resolved Hide resolved

Don't create an intermediate place when reading discriminants

1495b6b

Note to self

92e4471

oli-obk added 5 commits February 15, 2019 20:45

Insert satisfying op_to_const bbqing sound here.

0075868

Make validity checking use MPlaceTy instead of OpTy

a13cfdb

Eliminate another use of lazy_const_to_op

981a79e

Simplify raw_const_to_mplace

c07162c

Update a comment

defa8d5

Another bad function bites the dust

b653dc2

oli-obk added 3 commits February 15, 2019 21:58

Update an outdated comment

d3d145e

Keep MPlaceTy internals private

1f718b0

Don't access the stack if not strictly necessary

5ef183d

RalfJung reviewed Feb 16, 2019

View reviewed changes

src/librustc_mir/interpret/step.rs Show resolved Hide resolved

RalfJung reviewed Feb 16, 2019

View reviewed changes

oli-obk mentioned this pull request Feb 16, 2019

Const to op simplification #58511

Merged

RalfJung reviewed Feb 16, 2019

View reviewed changes

Centril added a commit to Centril/rust that referenced this pull request Feb 20, 2019

Rollup merge of rust-lang#58511 - oli-obk:const_to_op, r=RalfJung

a8722c2

Const to op simplification r? @RalfJung alternative to rust-lang#58486

Centril added a commit to Centril/rust that referenced this pull request Feb 23, 2019

Rollup merge of rust-lang#58511 - oli-obk:const_to_op, r=RalfJung

89c4253

Const to op simplification r? @RalfJung alternative to rust-lang#58486

Centril added a commit to Centril/rust that referenced this pull request Feb 24, 2019

Rollup merge of rust-lang#58511 - oli-obk:const_to_op, r=RalfJung

284787b

Const to op simplification r? @RalfJung alternative to rust-lang#58486

Centril added a commit to Centril/rust that referenced this pull request Feb 24, 2019

Rollup merge of rust-lang#58511 - oli-obk:const_to_op, r=RalfJung

1082a29

Const to op simplification r? @RalfJung alternative to rust-lang#58486

oli-obk closed this Feb 25, 2019

oli-obk deleted the import_miri_from_future branch March 16, 2021 12:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

always back `ty::Const` by an `Allocation` #58486

always back `ty::Const` by an `Allocation` #58486

oli-obk commented Feb 15, 2019

This comment has been minimized.

RalfJung Feb 15, 2019

oli-obk commented Feb 15, 2019

bors commented Feb 15, 2019

bors commented Feb 15, 2019

oli-obk commented Feb 15, 2019

rust-timer commented Feb 15, 2019

rust-timer commented Feb 15, 2019

RalfJung commented Feb 15, 2019

RalfJung Feb 16, 2019

RalfJung Feb 16, 2019

RalfJung Feb 16, 2019

RalfJung Feb 16, 2019

oli-obk Feb 16, 2019

bors commented Feb 24, 2019

RalfJung commented Feb 24, 2019

always back ty::Const by an Allocation #58486

always back ty::Const by an Allocation #58486

Conversation

oli-obk commented Feb 15, 2019

This comment has been minimized.

RalfJung Feb 15, 2019

Choose a reason for hiding this comment

oli-obk commented Feb 15, 2019

bors commented Feb 15, 2019

bors commented Feb 15, 2019

oli-obk commented Feb 15, 2019

rust-timer commented Feb 15, 2019

rust-timer commented Feb 15, 2019

RalfJung commented Feb 15, 2019

RalfJung Feb 16, 2019

Choose a reason for hiding this comment

RalfJung Feb 16, 2019

Choose a reason for hiding this comment

RalfJung Feb 16, 2019

Choose a reason for hiding this comment

RalfJung Feb 16, 2019

Choose a reason for hiding this comment

oli-obk Feb 16, 2019

Choose a reason for hiding this comment

bors commented Feb 24, 2019

RalfJung commented Feb 24, 2019

always back `ty::Const` by an `Allocation` #58486

always back `ty::Const` by an `Allocation` #58486