Initially MIR differentiated between arguments and locals, which
introduced a need to add extra copies assigning the argument to a
local, even for simple bindings. This differentiation no longer exists,
but we're still creating those copies, bloating the MIR and LLVM IR we
emit.
Additionally, the current approach means that we create debug info for
both the incoming argument (marking it as an argument), and then
immediately shadow it a local that goes by the same name. This can be
confusing when using e.g. "info args" in gdb, or when e.g. a debugger
with a GUI displays the function arguments separately from the local
variables, especially when the binding is mutable, because the argument
doesn't change, while the local variable does.
The extra alloca was only necessary because it made LLVM implicitly
handle the necessary deref to get to the actual value. The same happens
for indirect arguments that have the byval attribute. But the Rust ABI
does not use the byval attribute and so we need to manually add the
deref operation to the debuginfo.
This commit refactors the `collect_crate_translation_items` function to only
require the `TyCtxt` instead of a `SharedCrateContext` in preparation for
query-ifying this portion of trans.
Previously a `Symbol` was stored there, but this ended up causing hash
collisions in situations that otherwise shouldn't have a hash collision. Only
the symbol's string value was hashed, but it was possible for distinct symbols
to have the same string value, fooling various calcuations into thinking that
these paths *didn't* need disambiguating data when in fact they did!
By storing `InternedString` instead we're hopefully triggering all the exising
logic to disambiguate paths with same-name `Symbol` but actually distinct
locations.
This commit moves the calculation of the `LanguageItems` structure into a
query rather than being calculated before the `TyCtxt` exists, with the eventual
end goal of removing some `CrateStore` methods.
- fixes evaluation of array length for zero-sized type referenced by
rvalue operand.
- adds test to verify fix.
Cause of the issue.
Zero-sized aggregates are handled as operands, not lvalues. Therefore while
visiting Assign statement by LocalAnalyser, mark_as_lvalue() is not called for
related Local. This behaviour is controlled by rvalue_creates_operand() method.
As result it causes error later, when rvalue operand is evaluated in
trans_rvalue_operand() while handling Rvalue::Len case. Array length evaluation
invokes trans_lvalue() which expects referenced Local to be value, not operand.
How it is fixed.
In certain cases result of Rvalue::Len can be evaluated without calling
trans_lvalue(). Method evaluate_array_len() is introduced to handle length
evaluation for zero-sized types referenced by Locals.
Optimize initialization of arrays using repeat expressions
This PR was inspired by [this thread](https://www.reddit.com/r/rust/comments/6o8ok9/understanding_rust_performances_a_newbie_question/) on Reddit.
It tries to bring array initialization in the same ballpark as `Vec::from_elem()` for unoptimized builds.
For optimized builds this should relieve LLVM of having to figure out the construct we generate is in fact a `memset()`.
To that end this emits `llvm.memset()` when:
* the array is of integer type and all elements are zero (`Vec::from_elem()` also explicitly optimizes for this case)
* the array elements are byte sized
If the array is zero-sized initialization is omitted entirely.
APFloat: Rewrite It In Rust and use it for deterministic floating-point CTFE.
As part of the CTFE initiative, we're forced to find a solution for floating-point operations.
By design, IEEE-754 does not explicitly define everything in a deterministic manner, and there is some variability between platforms, at the very least (e.g. NaN payloads).
If types are to evaluate constant expressions involving type (or in the future, const) generics, that evaluation needs to be *fully deterministic*, even across `rustc` host platforms.
That is, if `[T; T::X]` was used in a cross-compiled library, and the evaluation of `T::X` executed a floating-point operation, that operation has to be reproducible on *any other host*, only knowing `T` and the definition of the `X` associated const (as either AST or HIR).
Failure to uphold those rules allows an associated type (e.g. `<Foo as Iterator>::Item`) to be seen as two (or more) different types, depending on the current host, and such type safety violations typically allow writing of a `transmute` in safe code, given enough generics.
The options considered by @rust-lang/compiler were:
1. Ban floating-point operations in generic const-evaluation contexts
2. Emulate floating-point operations in an uniformly deterministic fashion
The former option may seem appealing at first, but floating-point operations *are allowed today*, so they can't be banned wholesale, a distinction has to be made between the code that already works, and future generic contexts. *Moreover*, every computation that succeeded *has to be cached*, otherwise the generic case can be reproduced without any generics. IMO there are too many ways it can go wrong, and a single violation can be enough for an unsoundness hole.
Not to mention we may end up really wanting floating-point operations *anyway*, in CTFE.
I went with the latter option, and seeing how LLVM *already* has a library for this exact purpose (as it needs to perform optimizations independently of host floating-point capabilities), i.e. `APFloat`, that was what I ended up basing this PR on.
But having been burned by the low reusability of bindings that link to LLVM, and because I would *rather* the floating-point operations to be wrong than not deterministic or not memory-safe (`APFloat` does far more pointer juggling than I'm comfortable with), I decided to RIIR.
This way, we have a guarantee of *no* `unsafe` code, a bit more control over the where native floating-point might accidentally be involved, and non-LLVM backends can share it.
I've also ported all the testcases over, *before* any functionality, to catch any mistakes.
Currently the PR replaces all CTFE operations to go through `apfloat::ieee::{Single,Double}`, keeping only the bits of the `f32` / `f64` memory representation in between operations.
Converting from a string also double-checks that `core::num` and `apfloat` agree on the interpretation of a floating-point number literal, in case either of them has any bugs left around.
r? @nikomatsakis
f? @nagisa @est31
<hr/>
Huge thanks to @edef1c for first demoing usable `APFloat` bindings and to @chandlerc for fielding my questions on IRC about `APFloat` peculiarities (also upstreaming some bugfixes).
Add MIR Validate statement
This adds statements to MIR that express when types are to be validated (following [Types as Contracts](https://internals.rust-lang.org/t/types-as-contracts/5562)). Obviously nothing is stabilized, and in fact a `-Z` flag has to be passed for behavior to even change at all.
This is meant to make experimentation with Types as Contracts in miri possible. The design is definitely not final.
Cc @nikomatsakis @aturon
This elides initialization for zero-sized arrays:
* for zero-sized elements we previously emitted an empty loop
* for arrays with a length of zero we previously emitted a loop with zero
iterations
This emits llvm.memset() instead of a loop over each element when:
* all elements are zero integers
* elements are byte sized
The unwind path is always cold, so that should not have bad performance
implications. This avoids catastrophic exponential inlining, and also
decreases the size of librustc.so by 1.5% (OTOH, the size of `libstd.so`
increased by 0.5% for some reason).
Fixes#41696.
* Emit `EndRegion` for every code-extent for which we observe a
borrow. To do this, we needed to thread source info back through
to `fn in_scope`, which makes this commit a bit more painful than
one might have expected.
* There is `end_region` emission in `Builder::pop_scope` and in
`Builder::exit_scope`; the first handles falling out of a scope
normally, the second handles e.g. `break`.
* Remove `EndRegion` statements during the erase_regions mir
transformation.
* Preallocate the terminator block, and throw an `Unreachable` marker
on it from the outset. Then overwrite that Terminator as necessary
on demand.
* Instead of marking the scope as needs_cleanup after seeing a
borrow, just treat every scope in the chain as being part of the
diverge_block (after any *one* of them has separately signalled
that it needs cleanup, e.g. due to having a destructor to run).
* Allow for resume terminators to be patched when looking up drop flags.
(In particular, `MirPatch::new` has an explicit code path,
presumably previously unreachable, that patches up such resume
terminators.)
* Make `Scope` implement `Debug` trait.
* Expanded a stray comment: we do not emit StorageDead on diverging
paths, but that end behavior might not be desirable.
Translate array drop glue using MIR
I was a bit lazy here and used a usize-based index instead of a pointer iteration. Do you think this is important @eddyb?
r? @eddyb
