LLVM assumes that a thread will eventually cause side effect. This is
not true in Rust if a loop or recursion does nothing in its body,
causing undefined behavior even in common cases like `loop {}`.
Inserting llvm.sideeffect fixes the undefined behavior.
As a micro-optimization, only insert llvm.sideeffect when jumping back
in blocks or calling a function.
A patch for LLVM is expected to allow empty non-terminate code by
default and fix this issue from LLVM side.
https://github.com/rust-lang/rust/issues/28728
Remove tx_to_llvm_workers from TyCtxt
This can be kept within the codegen backend crates entirely -- there's no reason for us to create it outside and attempt to hold it in the (global) context.
Changes here aren't really too easily reviewable I suspect -- not sure if they can be cleaned up by splitting into more commits though, it's just hard to reason about `Box<Any>` in general. If there are thoughts though I'd be happy to hear them.
The primary goal of this PR is to get rid of the field on `rustc_interface::Queries`.
The old struct tail functions did not deal with `<T as Trait>::A` and `impl
Trait`, at least not explicitly. (We didn't notice this bug before because it
is only exposed when the tail (post deep normalization) is not `Sized`, so it
was a rare case to deal with.)
For post type-checking (i.e. during codegen), there is now
`struct_tail_erasing_lifetimes` and `struct_lockstep_tails_erasing_lifetimes`,
which each take an additional `ParamEnv` argument to drive normalization.
For pre type-checking cases where normalization is not needed, there is
`struct_tail_without_normalization`. (Currently, the only instance of this is
`Expectation::rvalue_hint`.)
All of these new entrypoints work by calling out to common helper routines.
The helpers are parameterized over a closure that handles the normalization.
Refactor C FFI variadics to more closely match their C counterparts, and add Clone implementation
We had to make some changes to expose `va_copy` and `va_end` directly to users (mainly for C2Rust, but not exclusively):
- redefine the Rust variadic structures to more closely correspond to C: `VaList` now matches `va_list`, and `VaListImpl` matches `__va_list_tag`
- add `Clone` for `VaListImpl`
- add explicit `as_va_list()` conversion function from `VaListImpl` to `VaList`
- add deref coercion from `VaList` to `VaListImpl`
- add support for the `asmjs` target
All these changes were needed for use cases like:
```Rust
let mut ap2 = va_copy(ap);
vprintf(fmt, ap2);
va_end(&mut ap2);
```
add support for unchecked math
add compiler support for
```rust
/// Returns the result of an unchecked addition, resulting in
/// undefined behavior when `x + y > T::max_value()` or `x + y < T::min_value()`.
pub fn unchecked_add<T>(x: T, y: T) -> T;
/// Returns the result of an unchecked substraction, resulting in
/// undefined behavior when `x - y > T::max_value()` or `x - y < T::min_value()`.
pub fn unchecked_sub<T>(x: T, y: T) -> T;
/// Returns the result of an unchecked multiplication, resulting in
/// undefined behavior when `x * y > T::max_value()` or `x * y < T::min_value()`.
pub fn unchecked_mul<T>(x: T, y: T) -> T;
```
cc https://github.com/rust-lang/rfcs/issues/2508
Remove LLVM instruction stats and other (obsolete) codegen stats.
Both `-Z count_llvm_insns` and `-Z codegen-stats` are removed, as (AFAIK) they have been of little use in the last few years, especially after the transition to MIR->LLVM codegen.
Other than for the LLVM instruction counts, `-Z codegen-stats` has long been obsoleted anyway.
r? @nagisa cc @rust-lang/compiler
The commit should have changed comments as well.
At the time of writting, it passes the tidy and check tool.
Revisions asked by eddyb :
- Renamed of all the occurences of {visit/super}_mir
- Renamed test structures `CachedMir` to `Cached`
Fixing the missing import on `AggregateKind`
This commit separates metadata encoding (`tcx.encode_metadata`) from the
creation of the metadata module (which is now handled by
`write_compressed_metadata`, formerly `write_metadata`).
The metadata encoding now occurs slightly earlier in the pipeline, at
the very start of code generation within `start_codegen`.
Metadata *writing* still occurs near the end of compilation; that will
be moved forward in subsequent commits.
