Struct OpaqueThreadedExecutionResult 

pub struct OpaqueThreadedExecutionResult<I> { /* private fields */ }

ThreadedExecutionResult in the shape tail-called handlers return it in.

Threaded handler internally returns this rather than ThreadedExecutionResult so that the outcome travels in registers instead of through a hidden out-pointer where possible for performance reasons (primarily on x86-64 due to a limited number of usable GPRs in the ABI).

On x86-64 the lanes are a 256-bit vector, so producing one of these executes AVX instructions - see Self::platform_supported() and Self::new().

Implementations

impl<I> OpaqueThreadedExecutionResult<I>

where I: Instruction,

pub fn platform_supported() -> bool

Whether this platform can carry an outcome the way Self::new() does.

It cannot on an x86-64 CPU without AVX, where the lanes are a 256-bit vector: Rust targets x86-64-v1 by default, so a runtime check is necessary.

This is called once within ThreadedExecutableInstruction::execute_threaded().

pub unsafe fn new(result: ThreadedExecutionResult<I>) -> Self

Serialize an outcome into the shape handlers return.

Safety

Self::platform_supported() must return true.

pub fn into_result(self) -> ThreadedExecutionResult<I>

Deserialize what Self::new() produced.

The lanes only ever come from there, in this very crate, which is what makes the unknown-tag arm unreachable.

Trait Implementations

impl<I: Clone> Clone for OpaqueThreadedExecutionResult<I>

fn clone(&self) -> OpaqueThreadedExecutionResult<I>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<I: Copy> Copy for OpaqueThreadedExecutionResult<I>

impl<I: Debug> Debug for OpaqueThreadedExecutionResult<I>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.