Struct rustc_mir_transform::nrvo::RenameReturnPlace

pub struct RenameReturnPlace;

This pass looks for MIR that always copies the same local into the return place and eliminates the copy by renaming all uses of that local to _0.

This allows LLVM to perform an optimization similar to the named return value optimization (NRVO) that is guaranteed in C++. This avoids a stack allocation and memcpy for the relatively common pattern of allocating a buffer on the stack, mutating it, and returning it by value like so:

fn foo(init: fn(&mut [u8; 1024])) -> [u8; 1024] {
    let mut buf = [0; 1024];
    init(&mut buf);
    buf
}

For now, this pass is very simple and only capable of eliminating a single copy. A more general version of copy propagation, such as the one based on non-overlapping live ranges in #47954 and #71003, could yield even more benefits.

Trait Implementations

impl<'tcx> MirPass<'tcx> for RenameReturnPlace

fn is_enabled(&self, sess: &Session) -> bool

Returns true if this pass is enabled with the current combination of compiler flags.

fn run_pass(&self, tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>)

fn name(&self) -> Cow<'_, str>

fn phase_change(&self) -> Option<MirPhase>

If this pass causes the MIR to enter a new phase, return that phase.

fn is_mir_dump_enabled(&self) -> bool

Layout

Note: Most layout information is completely unstable and may even differ between compilations. The only exception is types with certain repr(...) attributes. Please see the Rust Reference’s “Type Layout” chapter for details on type layout guarantees.

Size: 0 bytes