Validate CopyForDeref and DerefTemps better and remove them from runtime MIR
(split from my WIP rust-lang/rust#145344)
This PR:
- Removes `Rvalue::CopyForDeref` and `LocalInfo::DerefTemp` from runtime MIR
- Using a new mir pass `EraseDerefTemps`
- `CopyForDeref(x)` is turned into `Use(Copy(x))`
- `DerefTemp` is turned into `Boring`
- Not sure if this part is actually necessary, it made more sense in rust-lang/rust#145344 with `DerefTemp` storing actual data that I wanted to keep from having to be kept in sync with the rest of the body in runtime MIR
- Checks in validation that `CopyForDeref` and `DerefTemp` are only used together
- Removes special handling for `CopyForDeref` from many places
- Removes `CopyForDeref` from `custom_mir` reverting rust-lang/rust#111587
- In runtime MIR simple copies can be used instead
- In post cleanup analysis MIR it was already wrong to use due to the lack of support for creating `DerefTemp` locals
- Possibly this should be its own PR?
- Adds an argument to `deref_finder` to avoid creating new `DerefTemp`s and `CopyForDeref` in runtime MIR.
- Ideally we would just avoid making intermediate derefs instead of fixing it at the end of a pass / during shim building
- Removes some usages of `deref_finder` that I found out don't actually do anything
r? oli-obk
Port the implemention of SIMD intrinsics from Miri to const-eval
Ported the implementation of most SIMD intrinsics from Miri to rustc_const_eval. Remaining are
- Math functions (as per `@RalfJung's` suggestions)
- FMA (non-deterministic)
- Funnel Shifts (not implemented in Miri yet)
- Unordered reduction intrinsics (not implemented in Miri yet)
Turn ProjectionElem::Subtype into CastKind::Subtype
I noticed that drop elaboration can't, in general, handle `ProjectionElem::SubType`. It creates a disjoint move path that overlaps with other move paths. (`Subslice` does too, and I'm working on a different PR to make that special case less fragile.) If its skipped and treated as the same move path as its parent then `MovePath.place` has multiple possible projections. (It would probably make sense to remove all `Subtype` projections for the canonical place but it doesn't make sense to have this special case for a problem that doesn't actually occur in real MIR.)
The only reason this doesn't break is that `Subtype` is always the sole projection of the local its applied to. For the same reason, it works fine as a `CastKind` so I figured that makes more sense than documenting and validating this hidden invariant.
cc rust-lang/rust#112651, rust-lang/rust#133258
r? Icnr (bc you've been the main person dealing with `Subtype` it looks like)
Much of the compiler calls functions on Align projected from AbiAlign.
AbiAlign impls Deref to its inner Align, so we can simplify these away.
Also, it will minimize disruption when AbiAlign is removed.
For now, preserve usages that might resolve to PartialOrd or PartialEq,
as those have odd inference.
Clean up `ty::Dynamic`
1. As a follow-up to PR rust-lang/rust#143036, remove `DynKind` entirely.
2. Inside HIR ty lowering, consolidate modules `dyn_compatibility` and `lint` into `dyn_trait`
* `dyn_compatibility` wasn't about dyn compatibility itself, it's about lowering trait object types
* `lint` contained dyn-Trait-specific diagnostics+lints only
interpret: copy_provenance: avoid large intermediate buffer for large repeat counts
Copying provenance worked in this odd way where the "preparation" phase (which is supposed to just extract the necessary information from the source range) already did all the work of repeating the result N times for the target range. This was needed to use the existing `insert_presorted` function on `SortedMap`.
This PR generalizes `insert_presorted` so that we can avoid this odd structure on copy-provenance, and maybe even improve performance.
interpret: fix overlapping aggregate initialization
This fixes the problem pointed out by ````@saethlin```` in https://github.com/rust-lang/rust/issues/146383#issuecomment-3273224645.
Also clarify when exactly current de-facto MIR semantics allow overlap of the LHS and RHS in an assignment.
Implement `#[rustc_align_static(N)]` on `static`s
Tracking issue: https://github.com/rust-lang/rust/issues/146177
```rust
#![feature(static_align)]
#[rustc_align_static(64)]
static SO_ALIGNED: u64 = 0;
```
We need a different attribute than `rustc_align` because unstable attributes are tied to their feature (we can't have two unstable features use the same unstable attribute). Otherwise this uses all of the same infrastructure as `#[rustc_align]`.
r? `@traviscross`
We need a different attribute than `rustc_align` because unstable attributes are
tied to their feature (we can't have two unstable features use the same
unstable attribute). Otherwise this uses all of the same infrastructure
as `#[rustc_align]`.
rename erase_regions to erase_and_anonymize_regions
I find it consistently confusing that `erase_regions` does more than replacing regions with `'erased`. it also makes some code look real goofy to be writing manual folders to erase regions with a comment saying "we cant use erase regions" :> or code that re-calls erase_regions on types with regions already erased just to anonymize all the bound regions.
r? lcnr
idk how i feel about the name being almost twice as long now
const-eval: disable pointer fragment support
This fixes https://github.com/rust-lang/rust/issues/146291 by disabling pointer fragment support for const-eval. I want to properly fix this eventually, but won't get to it in the next few weeks, so this is an emergency patch to prevent the buggy implementation from landing on stable. The beta cutoff is on Sep 12th so if this PR lands after that, we'll need a backport.
Miri: fix handling of in-place argument and return place handling
This fixes two separate bugs (in two separate commits):
- If the return place is `_local` and not `*ptr`, we didn't always properly protect it if there were other pointers pointing to that return place.
- If two in-place arguments are *the same* local variable, we didn't always detect that aliasing.
Add tracing to various miscellaneous functions
This PR adds tracing to:
- `ty.fn_sig()`. There is only one place where `fn_sig` is called for real within `rustc_const_eval`. There are three other places where it's called, but one is inside `ConstCx::fn_sig` (which does not seem to be used anywhere), another is under `if cfg!(debug_assertions)`, and the last is within `call_main` and thus gets called only once.
- the two possible things `find_mir_or_eval_fn` can do: "emulate_foreign_item" and "load_mir"
- all calls to `Const.eval()` within the Miri or the `rustc_const_eval` codebase.
- a separate commit also fixes the style of some tracing macros
Those are all quite long-lived operations, that in total make up for 6-7% of the total time spent in the program. I found out about them by looking for long periods of time that were previously not traced at all, using this SQL query in ui.perfetto.dev:
```sql
with ordered as (select s1.*, row_number() over (order by s1.ts) as rn from slices as s1 where s1.parent_id is null and s1.dur > 0 and s1.name != "frame" and s1.name != "step" and s1.name != "backtrace") select a.ts+a.dur as ts, b.ts-a.ts-a.dur as dur, a.id, a.track_id, a.category, a.depth, a.stack_id, a.parent_stack_id, a.parent_id, a.arg_set_id, a.thread_ts, a.thread_instruction_count, a.thread_instruction_delta, a.cat, a.slice_id, "empty" as name from ordered as a inner join ordered as b on a.rn=b.rn-1 /*where b.ts-a.ts-a.dur > 5000*/ order by b.ts-a.ts-a.dur desc
```
<details>
<summary>How the table was obtained</summary>
The above image was obtained in ui.perfetto.dev with the following SQL query after obtaining a trace file by running Miri on the following Rust code with `n=100`.
```sql
select "TOTAL PROGRAM DURATION" as name, count(*), max(ts + dur) as "sum(dur)", 100.0 as "%", null as "min(dur)", null as "max(dur)", null as "avg(dur)", null as "stddev(dur)" from slices union select "TOTAL OVER ALL SPANS (excluding events)" as name, count(*), sum(dur), cast(cast(sum(dur) as float) / (select max(ts + dur) from slices) * 1000 as int) / 10.0 as "%", min(dur), max(dur), cast(avg(dur) as int) as "avg(dur)", cast(sqrt(avg(dur*dur)-avg(dur)*avg(dur)) as int) as "stddev(dur)" from slices where parent_id is null and name != "frame" and name != "step" and dur > 0 union select name, count(*), sum(dur), cast(cast(sum(dur) as float) / (select max(ts + dur) from slices) * 1000 as int) / 10.0 as "%", min(dur), max(dur), cast(avg(dur) as int) as "avg(dur)", cast(sqrt(avg(dur*dur)-avg(dur)*avg(dur)) as int) as "stddev(dur)" from slices where parent_id is null and name != "frame" and name != "step" group by name order by sum(dur) desc, count(*) desc
```
```rust
fn main() {
let n: usize = std::env::args().nth(1).unwrap().parse().unwrap();
let mut v = (0..n).into_iter().collect::<Vec<_>>();
for i in &mut v {
*i += 1;
}
}
```
</details>
<img width="1689" height="317" alt="image" src="https://github.com/user-attachments/assets/ee2c81f5-d74a-4da5-b4b6-ab2770175b14" />
const-eval: full support for pointer fragments
This fixes https://github.com/rust-lang/const-eval/issues/72 and makes `swap_nonoverlapping` fully work in const-eval by enhancing per-byte provenance tracking with tracking of *which* of the bytes of the pointer this one is. Later, if we see all the same bytes in the exact same order, we can treat it like a whole pointer again without ever risking a leak of the data bytes (that encode the offset into the allocation). This lifts the limitation that was discussed quite a bit in https://github.com/rust-lang/rust/pull/137280.
For a concrete piece of code that used to fail and now works properly consider this example doing a byte-for-byte memcpy in const without using intrinsics:
```rust
use std::{mem::{self, MaybeUninit}, ptr};
type Byte = MaybeUninit<u8>;
const unsafe fn memcpy(dst: *mut Byte, src: *const Byte, n: usize) {
let mut i = 0;
while i < n {
*dst.add(i) = *src.add(i);
i += 1;
}
}
const _MEMCPY: () = unsafe {
let ptr = &42;
let mut ptr2 = ptr::null::<i32>();
// Copy from ptr to ptr2.
memcpy(&mut ptr2 as *mut _ as *mut _, &ptr as *const _ as *const _, mem::size_of::<&i32>());
assert!(*ptr2 == 42);
};
```
What makes this code tricky is that pointers are "opaque blobs" in const-eval, we cannot just let people look at the individual bytes since *we don't know what those bytes look like* -- that depends on the absolute address the pointed-to object will be placed at. The code above "breaks apart" a pointer into individual bytes, and then puts them back together in the same order elsewhere. This PR implements the logic to properly track how those individual bytes relate to the original pointer, and to recognize when they are in the right order again.
We still reject constants where the final value contains a not-fully-put-together pointer: I have no idea how one could construct an LLVM global where one byte is defined as "the 3rd byte of a pointer to that other global over there" -- and even if LLVM supports this somehow, we can leave implementing that to a future PR. It seems unlikely to me anyone would even want this, but who knows.^^
This also changes the behavior of Miri, by tracking the order of bytes with provenance and only considering a pointer to have valid provenance if all bytes are in the original order again. This is related to https://github.com/rust-lang/unsafe-code-guidelines/issues/558. It means one cannot implement XOR linked lists with strict provenance any more, which is however only of theoretical interest. Practically I am curious if anyone will show up with any code that Miri now complains about - that would be interesting data. Cc `@rust-lang/opsem`
Add tracing to resolve-related functions
Resolve-related functions are not called often but still make up for ~3% of execution time for non-repetitive programs (as seen in the first table below, obtained from running the rust snippet at the bottom with `n=1`). On the other hand, for repetitive programs they become less relevant (I tested the same snippet but with `n=100` and got ~1.5%), and it appears that only `try_resolve` is called more often (see the last two tables).
The first table was obtained by opening the trace file in https://ui.perfetto.dev and running the following query:
```sql
select "TOTAL PROGRAM DURATION" as name, count(*), max(ts + dur) as "sum(dur)", 100.0 as "%", null as "min(dur)", null as "max(dur)", null as "avg(dur)", null as "stddev(dur)" from slices union select "TOTAL OVER ALL SPANS (excluding events)" as name, count(*), sum(dur), cast(cast(sum(dur) as float) / (select max(ts + dur) from slices) * 1000 as int) / 10.0 as "%", min(dur), max(dur), cast(avg(dur) as int) as "avg(dur)", cast(sqrt(avg(dur*dur)-avg(dur)*avg(dur)) as int) as "stddev(dur)" from slices where parent_id is null and name != "frame" and name != "step" and dur > 0 union select name, count(*), sum(dur), cast(cast(sum(dur) as float) / (select max(ts + dur) from slices) * 1000 as int) / 10.0 as "%", min(dur), max(dur), cast(avg(dur) as int) as "avg(dur)", cast(sqrt(avg(dur*dur)-avg(dur)*avg(dur)) as int) as "stddev(dur)" from slices where parent_id is null and name != "frame" and name != "step" group by name order by sum(dur) desc, count(*) desc
```
<img width="1687" height="242" alt="image" src="https://github.com/user-attachments/assets/4d4bd890-869b-40f3-a473-8e4c42b02da4" />
The following two tables show how many `resolve` spans there per subname/subcategory, and how much time is spent in each. The first is for `n=1` and the second for `n=100`. The query that was used is:
```sql
select args.string_value as name, count(*), max(dur), avg(dur), sum(dur) from slices inner join args USING (arg_set_id) where args.key = "args." || slices.name and name = "resolve" group by args.string_value
```
<img width="1688" height="159" alt="image" src="https://github.com/user-attachments/assets/a8749856-c099-492e-a86e-6d67b146af9c" />
<img width="1688" height="159" alt="image" src="https://github.com/user-attachments/assets/ce3ac1b5-5c06-47d9-85a6-9b921aea348e" />
The snippet I tested with Miri to obtain the above traces is:
```rust
fn main() {
let n: usize = std::env::args().nth(1).unwrap().parse().unwrap();
let mut v = (0..n).into_iter().collect::<Vec<_>>();
for i in &mut v {
*i += 1;
}
}
```
`Printer` cleanups
The trait `Printer` is implemented by six types, and the sub-trait `PrettyPrinter` is implemented by three of those types. The traits and the impls are complex and a bit of a mess. This PR starts to clean them up.
r? ``@davidtwco``
Do not give function allocations alignment in consteval and Miri.
We do not yet have a (clear and T-lang approved) design for how `#[align(N)]` on functions should affect function pointers' addresses on various platforms, so for now do not give function pointers alignment in consteval and Miri.
----
Old summary:
Not a full solution to <https://github.com/rust-lang/rust/issues/144661>, but fixes the immediate issue by making function allocations all have alignment 1 in consteval, ignoring `#[rustc_align(N)]`, so the compiler doesn't know if any offset other than 0 is non-null.
A more "principlied" solution would probably be to make function pointers to `#[instruction_set(arm::t32)]` functions be at offset 1 of an align-`max(2, align attribute)` allocation instead of at offset 0 of their allocation during consteval, and on wasm to either disallow `#[align(N)]` where N > 1, or to pad the function table such that the function pointer of a `#[align(N)]` function is a multiple of `N` at runtime.
Currently they are mostly named `cx`, which is a terrible name for a
type that impls `Printer`/`PrettyPrinter`, and is easy to confuse with
other types like `TyCtxt`. This commit changes them to `p`. A couple of
existing `p` variables had to be renamed to make way.
