Add intrinsics `fmuladd{f16,f32,f64,f128}`. This computes `(a * b) +
c`, to be fused if the code generator determines that (i) the target
instruction set has support for a fused operation, and (ii) that the
fused operation is more efficient than the equivalent, separate pair
of `mul` and `add` instructions.
https://llvm.org/docs/LangRef.html#llvm-fmuladd-intrinsic
MIRI support is included for f32 and f64.
The codegen_cranelift uses the `fma` function from libc, which is a
correct implementation, but without the desired performance semantic. I
think this requires an update to cranelift to expose a suitable
instruction in its IR.
I have not tested with codegen_gcc, but it should behave the same
way (using `fma` from libc).
- fix for divergence
- fix error message
- fix another cranelift test
- fix some cranelift things
- don't set the NORETURN option for naked asm
- fix use of naked_asm! in doc comment
- fix use of naked_asm! in run-make test
- use `span_bug` in unreachable branch
simd_shuffle intrinsic: allow argument to be passed as vector
See https://github.com/rust-lang/rust/issues/128738 for context.
I'd like to get rid of [this hack](6c0b89dfac/compiler/rustc_codegen_ssa/src/mir/block.rs (L922-L935)). https://github.com/rust-lang/rust/pull/128537 almost lets us do that since constant SIMD vectors will then be passed as immediate arguments. However, simd_shuffle for some reason actually takes an *array* as argument, not a vector, so the hack is still required to ensure that the array becomes an immediate (which then later stages of codegen convert into a vector, as that's what LLVM needs).
This PR prepares simd_shuffle to also support a vector as the `idx` argument. Once this lands, stdarch can hopefully be updated to pass `idx` as a vector, and then support for arrays can be removed, which finally lets us get rid of that hack.
Shrink `TyKind::FnPtr`.
By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and `FnHeader`, which can be packed more efficiently. This reduces the size of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms. This reduces peak memory usage by a few percent on some benchmarks. It also reduces cache misses and page faults similarly, though this doesn't translate to clear cycles or wall-time improvements on CI.
r? `@compiler-errors`
nontemporal_store: make sure that the intrinsic is truly just a hint
The `!nontemporal` flag for stores in LLVM *sounds* like it is just a hint, but actually, it is not -- at least on x86, non-temporal stores need very special treatment by the programmer or else the Rust memory model breaks down. LLVM still treats these stores as-if they were normal stores for optimizations, which is [highly dubious](https://github.com/llvm/llvm-project/issues/64521). Let's avoid all that dubiousness by making our own non-temporal stores be truly just a hint, which is possible on some targets (e.g. ARM). On all other targets, non-temporal stores become regular stores.
~~Blocked on https://github.com/rust-lang/stdarch/pull/1541 propagating to the rustc repo, to make sure the `_mm_stream` intrinsics are unaffected by this change.~~
Fixes https://github.com/rust-lang/rust/issues/114582
Cc `@Amanieu` `@workingjubilee`
By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and
`FnHeader`, which can be packed more efficiently. This reduces the size
of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms.
This reduces peak memory usage by a few percent on some benchmarks. It
also reduces cache misses and page faults similarly, though this doesn't
translate to clear cycles or wall-time improvements on CI.
Add a hook for `should_codegen_locally`
This PR lifts the module-local function `should_codegen_locally` to `TyCtxt` as a hook.
In addition to monomorphization, this function is used for checking the dependency of `compiler_builtins` on other libraries. Moving this function to the hooks also makes overriding it possible for the tools that use the rustc interface.
Clean up a few minor refs in `format!` macro, as it has a performance cost. Apparently the compiler is unable to inline `format!("{}", &variable)`, and does a run-time double-reference instead (format macro already does one level referencing). Inlining format args prevents accidental `&` misuse.
Sync ar_archive_writer to LLVM 18.1.3
From LLVM 15.0.0-rc3. This adds support for COFF archives containing Arm64EC object files and has various fixes for AIX big archive files.
Support tail calls in mir via `TerminatorKind::TailCall`
This is one of the interesting bits in tail call implementation — MIR support.
This adds a new `TerminatorKind` which represents a tail call:
```rust
TailCall {
func: Operand<'tcx>,
args: Vec<Operand<'tcx>>,
fn_span: Span,
},
```
*Structurally* this is very similar to a normal `Call` but is missing a few fields:
- `destination` — tail calls don't write to destination, instead they pass caller's destination to the callee (such that eventual `return` will write to the caller of the function that used tail call)
- `target` — similarly to `destination` tail calls pass the caller's return address to the callee, so there is nothing to do
- `unwind` — I _think_ this is applicable too, although it's a bit confusing
- `call_source` — `become` forbids operators and is not created as a lowering of something else; tail calls always come from HIR (at least for now)
It might be helpful to read the interpreter implementation to understand what `TailCall` means exactly, although I've tried documenting it too.
-----
There are a few `FIXME`-questions still left, ideally we'd be able to answer them during review ':)
-----
r? `@oli-obk`
cc `@scottmcm` `@DrMeepster` `@JakobDegen`
Miri function identity hack: account for possible inlining
Having a non-lifetime generic is not the only reason a function can be duplicated. Another possibility is that the function may be eligible for cross-crate inlining. So also take into account the inlining attribute in this Miri hack for function pointer identity.
That said, `cross_crate_inlinable` will still sometimes return true even for `inline(never)` functions:
- when they are `DefKind::Ctor(..) | DefKind::Closure` -- I assume those cannot be `InlineAttr::Never` anyway?
- when `cross_crate_inline_threshold == InliningThreshold::Always`
so maybe this is still not quite the right criterion to use for function pointer identity.
