When overflow checking on `<<` and `>>` was added for integers, the `<<` and `>>` operations broke for SIMD types (`u32x4`, `i16x8`, etc.). This PR implements checked shifts on SIMD types.
Fixes#24258.
This has a number of advantages compared to creating a copy in memory
and passing a pointer. The obvious one is that we don't have to put the
data into memory but can keep it in registers. Since we're currently
passing a pointer anyway (instead of using e.g. a known offset on the
stack, which is what the `byval` attribute would achieve), we only use a
single additional register for each fat pointer, but save at least two
pointers worth of stack in exchange (sometimes more because more than
one copy gets eliminated). On archs that pass arguments on the stack, we
save a pointer worth of stack even without considering the omitted
copies.
Additionally, LLVM can optimize the code a lot better, to a large degree
due to the fact that lots of copies are gone or can be optimized away.
Additionally, we can now emit attributes like nonnull on the data and/or
vtable pointers contained in the fat pointer, potentially allowing for
even more optimizations.
This results in LLVM passes being about 3-7% faster (depending on the
crate), and the resulting code is also a few percent smaller, for
example:
text data filename
5671479 3941461 before/librustc-d8ace771.so
5447663 3905745 after/librustc-d8ace771.so
1944425 2394024 before/libstd-d8ace771.so
1896769 2387610 after/libstd-d8ace771.so
I had to remove a call in the backtrace-debuginfo test, because LLVM can
now merge the tails of some blocks when optimizations are turned on,
which can't correctly preserve line info.
Fixes#22924
Cc #22891 (at least for fat pointers the code is good now)
When taking the address of an unsized field we generate a rvalue datum
for the field and then convert it to an lvalue datum. At that point,
cleanup is scheduled for the field, leading to multiple drop calls.
The problem is that we generate an rvalue datum for the field, since the
pointer does not own the data and there's already cleanup scheduled
elsewhere by the true owner. Instead, an lvalue datum must be created.
Thanks to @eddyb for identifying the underlying cause and suggesting the
correct fix.
Fixes#25549.
I think I didn't run tests properly - my second call to
select_all_obligations_or_error has made 3 tests fail. However, this is
just an error message change - integer fallback never worked with casts.
This should hopefully fix all cast-related ICEs once and for all.
I managed to make diagnostics hate me and give me spurious "decoder error"
- removing $build/tmp/extended-errors seems to fix it.
When taking the address of an unsized field we generate a rvalue datum
for the field and then convert it to an lvalue datum. At that point,
cleanup is scheduled for the field, leading to multiple drop calls.
The problem is that we generate an rvalue datum for the field, since the
pointer does not own the data and there's already cleanup scheduled
elsewhere by the true owner. Instead, an lvalue datum must be created.
Thanks to @eddyb for identifying the underlying cause and suggesting the
correct fix.
Fixes#25549Fixes#25515
* segfault due to not copying drop flag when coercing
* fat pointer casts
* segfault due to not checking drop flag properly
* debuginfo for DST smart pointers
* unreachable code in drop glue
The former stopped making sense when we started interning substs and made
TraitRef a 2-word copy type, and I'm moving the latter into an arena as
they live as long as the type context.
the plan is to treat all binary operators as if they were overloaded,
relying on the fact that we have impls for all the builtin scalar
operations (and no more). But then during writeback we clear the
overload if the types correspond to a builtin op.
This strategy allows us to avoid having to know the types of the
operands ahead of time. It also avoids us overspecializing as we did in
the past.
This permits all coercions to be performed in casts, but adds lints to warn in those cases.
Part of this patch moves cast checking to a later stage of type checking. We acquire obligations to check casts as part of type checking where we previously checked them. Once we have type checked a function or module, then we check any cast obligations which have been acquired. That means we have more type information available to check casts (this was crucial to making coercions work properly in place of some casts), but it means that casts cannot feed input into type inference.
[breaking change]
* Adds two new lints for trivial casts and trivial numeric casts, these are warn by default, but can cause errors if you build with warnings as errors. Previously, trivial numeric casts and casts to trait objects were allowed.
* The unused casts lint has gone.
* Interactions between casting and type inference have changed in subtle ways. Two ways this might manifest are:
- You may need to 'direct' casts more with extra type information, for example, in some cases where `foo as _ as T` succeeded, you may now need to specify the type for `_`
- Casts do not influence inference of integer types. E.g., the following used to type check:
```
let x = 42;
let y = &x as *const u32;
```
Because the cast would inform inference that `x` must have type `u32`. This no longer applies and the compiler will fallback to `i32` for `x` and thus there will be a type error in the cast. The solution is to add more type information:
```
let x: u32 = 42;
let y = &x as *const u32;
```
This is a [breaking-change]. When indexing a generic map (hashmap, etc) using the `[]` operator, it is now necessary to borrow explicitly, so change `map[key]` to `map[&key]` (consistent with the `get` routine). However, indexing of string-valued maps with constant strings can now be written `map["abc"]`.
r? @japaric
cc @aturon @Gankro
After this patch code like `let ref a = *"abcdef"` doesn't cause ICE anymore.
Required for #23121
There are still places in rustc_trans where pointers are always assumed to be thin. In particular, #19064 is not resolved by this patch.
This includes a slight refactoring of the `cast_shift_rhs` and related
functions in `trans::base`, so that I can call them from much later in
the compiler's control flow (so that we can clearly dilineate where
automatic conversions of the RHS occur, versus where we check it).
The rhs-checking and fallback-masking is generalized to 8- and 16-bit
values, and the fallback-masking is turned on unconditionally.
Fix#10183.
Is this a [breaking-change]? I would argue it is not; it only adds a
strict definition to what was previously undefined behavior; however,
there might be code that was e.g. assuming that `1_i8 << 17` yields 0.
(This happens in certain contexts and at certain optimization levels.)
Safe fns are no longer subtypes of unsafe fns, but you can coerce from one to the other.
This is a [breaking-change] in that impl fns must now be declared `unsafe` if the trait is declared `unsafe`. In some rare cases, the subtyping change may also direct affect you, but no such cases were encountered in practice.
Fixes#23449.
r? @nrc
This upcast coercion currently preserves the vtable for the object, but
eventually it can be used to create a derived vtable. The upcast
coercion is not introduced into method dispatch; see comment on #18737
for information about why. Fixes#18737.
Adds overflow checking to integer addition, multiplication, and subtraction
when `-Z force-overflow-checks` is true, or if `--cfg ndebug` is not passed to
the compiler. On overflow, it panics with `arithmetic operation overflowed`.
Also adds `overflowing_add`, `overflowing_sub`, and `overflowing_mul`
intrinsics for doing unchecked arithmetic.
[breaking-change]
Take 2. This PR includes a bunch of refactoring that was part of an experimental branch implementing [implied bounds]. That particular idea isn't ready to go yet, but the refactoring proved useful for fixing #22246. The implied bounds branch also exposed #22110 so a simple fix for that is included here. I still think some more refactoring would be a good idea here -- in particular I think most of the code in wf.rs is kind of duplicating the logic in implicator and should go, but I decided to post this PR and call it a day before diving into that. I'll write a bit more details about the solutions I adopted in the various bugs. I patched the two issues I was concerned about, which was the handling of supertraits and HRTB (the latter turned out to be fine, so I added a comment explaining why.)
r? @pnkfelix (for now, anyway)
cc @aturon
[implied bounds]: http://smallcultfollowing.com/babysteps/blog/2014/07/06/implied-bounds/
This is super black magic internals at the moment, but having it
somewhere semi-public seems good. The current versions weren't being
rendered, and they'll be useful for some people.
Fixes#21281
This *almost* completes the job for #16440. The idea is that even if we do not know whether some closure type `C` implements `Fn` or `FnMut` (etc), we still know its argument and return types. So if we see an obligation `C : Fn(_0)`, we can unify `_0` with those argument types while still considering the obligation ambiguous and unsatisfied. This helps to make a lot of progress with type inference even before closure kind inference is done.
As part of this PR, the explicit `:` syntax is removed from the AST and completely ignored. We still infer the closure kind based on the expected type if that is available. There are several reasons for this. First, deciding the closure kind earlier is always better, as it allows us to make more progress. Second, this retains a (admittedly obscure) way for users to manually specify the closure kind, which is useful for writing tests if nothing else. Finally, there are still some cases where inference can fail, so it may be useful to have this manual override. (The expectation is that we will eventually revisit an explicit syntax for specifying the closure kind, but it will not be `:` and may be some sort of generalization of the `||` syntax to handle other traits as well.)
This commit does not *quite* fix#16640 because a snapshot is still needed to enable the obsolete syntax errors for explicit `&mut:` and friends.
r? @eddyb as he reviewed the prior patch in this direction
upgrade the inference based on expected type so that it is able to
infer the fn kind in isolation even if the full signature is not
available (and we could perhaps do better still in some cases, such as
extracting just the types of the arguments but not the return value).
trans: When coercing to `Box<Trait>` or `Box<[T]>`, leave datum in it's original L-/R-value state.
This fixes a subtle issue where temporaries were being allocated (but not necessarily initialized) to the (parent) terminating scope of a match expression; in particular, the code to zero out the temporary emitted by `datum.store_to` is only attached to the particular match-arm for that temporary, but when going down other arms of the match expression, the temporary may falsely appear to have been initialized, depending on what the stack held at that location, and thus may have its destructor erroneously run at the end of the terminating scope.
FIx#20055.
(There may be a latent bug still remaining in `fn into_fat_ptr`, but I am so annoyed by the test/run-pass/coerce_match.rs failures that I want to land this now.)
This fixes a subtle issue where temporaries were being allocated (but
not necessarily initialized) to the (parent) terminating scope of a
match expression; in particular, the code to zero out the temporary
emitted by `datum.store_to` is only attached to the particular
match-arm for that temporary, but when going down other arms of the
match expression, the temporary may falsely appear to have been
initialized, depending on what the stack held at that location, and
thus may have its destructor erroneously run at the end of the
terminating scope.
Test cases to appear in a follow-up commit.
Fix#20055
So far, the source location an LLVM instruction was linked to was controlled by
`debuginfo::set_source_location()` and `debuginfo::clear_source_location()`.
This interface mimicked how LLVM's `IRBuilder` handles debug location
assignment. While this interface has some theoretical performance benefits, it
also makes things terribly unstable: One sets some quasi-global state and then
hopes that it is still correct when a given instruction is emitted---an
assumption that has been proven to not hold a bit too often.
This patch requires the debug source location to be passed to the actual
instruction emitting function. This makes source location assignment explicit
and will prevent future changes to `trans` from accidentally breaking things in
the majority of cases.
This patch does not yet implement the new principle for all instruction kinds
but the stepping experience should have improved significantly nonetheless
already.
This stops the compiler ICEing on the use of SIMD types in FFI signatures. It emits correct code for LLVM intrinsics, but I am quite unsure about the ABI handling in general so I've added a new feature gate `simd_ffi` to try to ensure people don't use it without realising there's a non-trivial risk of codegen brokenness.
Closes#20043.
There are two places left where we used to only know the byte
size of/offset into an array and had to cast to i8 and back to get the
right addresses. But by now, we always know the sizes in terms of the
number of elements in the array. In fact we have to add an extra Mul
instruction so we can use the weird cast-to-u8 code. So we should really
just embrace our new knowledge and use simple GEPs to do the address
calculations.
Fixes#3729
There are two places left where we used to only know the byte
size of/offset into an array and had to cast to i8 and back to get the
right addresses. But by now, we always know the sizes in terms of the
number of elements in the array. In fact we have to add an extra Mul
instruction so we can use the weird cast-to-u8 code. So we should really
just embrace our new knowledge and use simple GEPs to do the address
calculations.
Additionally, the pointer calculations in bind_subslice_pat don't handle
zero-sized types correctly, producing slices that point outside the
array that is being matched against. Using GEP fixes that as well.
Fixes#3729
fmt::Show is for debugging, and can and should be implemented for
all public types. This trait is used with `{:?}` syntax. There still
exists #[derive(Show)].
fmt::String is for types that faithfully be represented as a String.
Because of this, there is no way to derive fmt::String, all
implementations must be purposeful. It is used by the default format
syntax, `{}`.
This will break most instances of `{}`, since that now requires the type
to impl fmt::String. In most cases, replacing `{}` with `{:?}` is the
correct fix. Types that were being printed specifically for users should
receive a fmt::String implementation to fix this.
Part of #20013
[breaking-change]
[breaking-change]
The `mut` in slices is now redundant. Mutability is 'inferred' from position. This means that if mutability is only obvious from the type, you will need to use explicit calls to the slicing methods.
#16081 fixed an issue where a nested return statement would cause incorrect behaviour due to the inner return writing over the return stack slot that had already been written too. However, the check was very broad and picked many cases that wouldn't ever be affected by this issue.
As a result, the number of allocas increased dramatically and therefore stack-size increased. LLVM is not able to remove all of the extraneous allocas. Any code that had multiple return values in a compound expression at the end of a function (including loops) would be hit by the issue.
The check now uses a control-flow graph to only consider the case when the inner return is executed conditionally. By itself, this narrowed definition causes #15763 to return, so the control-flow graph is also used to avoid passing the return slot as a destination when the result won't be used.
This change allows the stack-size of the main rustc task to be reduced to 8MB from 32MB.
- The following operator traits now take their arguments by value: `Add`, `Sub`, `Mul`, `Div`, `Rem`, `BitAnd`, `BitOr`, `BitXor`, `Shl`, `Shr`. This breaks all existing implementations of these traits.
- The binary operation `a OP b` now "desugars" to `OpTrait::op_method(a, b)` and consumes both arguments.
- `String` and `Vec` addition have been changed to reuse the LHS owned value, and to avoid internal cloning. Only the following asymmetric operations are available: `String + &str` and `Vec<T> + &[T]`, which are now a short-hand for the "append" operation.
[breaking-change]
---
This passes `make check` locally. I haven't touch the unary operators in this PR, but converting them to by value should be very similar to this PR. I can work on them after this gets the thumbs up.
@nikomatsakis r? the compiler changes
@aturon r? the library changes. I think the only controversial bit is the semantic change of the `Vec`/`String` `Add` implementation.
cc #19148
This change makes the compiler no longer infer whether types (structures
and enumerations) implement the `Copy` trait (and thus are implicitly
copyable). Rather, you must implement `Copy` yourself via `impl Copy for
MyType {}`.
A new warning has been added, `missing_copy_implementations`, to warn
you if a non-generic public type has been added that could have
implemented `Copy` but didn't.
For convenience, you may *temporarily* opt out of this behavior by using
`#![feature(opt_out_copy)]`. Note though that this feature gate will never be
accepted and will be removed by the time that 1.0 is released, so you should
transition your code away from using it.
This breaks code like:
#[deriving(Show)]
struct Point2D {
x: int,
y: int,
}
fn main() {
let mypoint = Point2D {
x: 1,
y: 1,
};
let otherpoint = mypoint;
println!("{}{}", mypoint, otherpoint);
}
Change this code to:
#[deriving(Show)]
struct Point2D {
x: int,
y: int,
}
impl Copy for Point2D {}
fn main() {
let mypoint = Point2D {
x: 1,
y: 1,
};
let otherpoint = mypoint;
println!("{}{}", mypoint, otherpoint);
}
This is the backwards-incompatible part of #13231.
Part of RFC #3.
[breaking-change]
One negative side-effect of this change is that there might be quite a bit of copying strings out of the codemap, i.e. one copy for every block that gets translated, just for taking a look at the last character of the block. If this turns out to cause a performance problem then `CodeMap::span_to_snippet()` could be changed return `Option<&str>` instead of `Option<String>`.
Fixes#18791
