Enum FastDivmod

pub enum FastDivmod {
    Fast {
        divisor: u32,
        multiplier: u32,
        shift_right: u32,
    },
    PowerOfTwo {
        shift: u32,
        mask: u32,
    },
    Fallback {
        divisor: u32,
    },
}

Create a fast-divmod object if supported, or a regular fallback if not. This precalculates certain values on the host, in exchange for making division and modulo operations on the GPU much faster. Only supports u32 right now to allow for a simpler algorithm. It’s mostly used for indices regardless.

Implementation based on ONNX: https://github.com/microsoft/onnxruntime/blob/main/onnxruntime/core/providers/cuda/shared_inc/fast_divmod.h

Variants

Fast

Fields

divisor: u32

multiplier: u32

shift_right: u32

PowerOfTwo

Fields

shift: u32

mask: u32

Fallback

Fields

divisor: u32

Implementations

impl FastDivmod

pub fn new_Fast(divisor: u32, multiplier: u32, shift_right: u32) -> Self

pub fn __expand_new_Fast( _: &mut Scope, divisor: <u32 as CubeType>::ExpandType, multiplier: <u32 as CubeType>::ExpandType, shift_right: <u32 as CubeType>::ExpandType, ) -> FastDivmodExpand

pub fn new_PowerOfTwo(shift: u32, mask: u32) -> Self

pub fn __expand_new_PowerOfTwo( _: &mut Scope, shift: <u32 as CubeType>::ExpandType, mask: <u32 as CubeType>::ExpandType, ) -> FastDivmodExpand

pub fn new_Fallback(divisor: u32) -> Self

pub fn __expand_new_Fallback( _: &mut Scope, divisor: <u32 as CubeType>::ExpandType, ) -> FastDivmodExpand

impl FastDivmod

pub fn div(&self, dividend: u32) -> u32

pub fn modulo(&self, dividend: u32) -> u32

pub fn div_mod(&self, dividend: u32) -> (u32, u32)

pub fn __expand_div( scope: &mut Scope, this: <Self as CubeType>::ExpandType, dividend: <u32 as CubeType>::ExpandType, ) -> <u32 as CubeType>::ExpandType

pub fn __expand_modulo( scope: &mut Scope, this: <Self as CubeType>::ExpandType, dividend: <u32 as CubeType>::ExpandType, ) -> <u32 as CubeType>::ExpandType

pub fn __expand_div_mod( scope: &mut Scope, this: <Self as CubeType>::ExpandType, dividend: <u32 as CubeType>::ExpandType, ) -> <(u32, u32) as CubeType>::ExpandType

Trait Implementations

impl Clone for FastDivmod

fn clone(&self) -> FastDivmod

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl CubeDebug for FastDivmod

fn set_debug_name(&self, scope: &mut Scope, name: &'static str)

Set the debug name of this type’s expansion. Should do nothing for types that don’t appear at runtime

impl CubeType for FastDivmod

type ExpandType = FastDivmodExpand

fn into_mut(scope: &mut Scope, expand: Self::ExpandType) -> Self::ExpandType

Wrapper around the init method, necessary to type inference.

impl LaunchArg for FastDivmod

type RuntimeArg<'a, R: Runtime> = FastDivmodArgs<'a, R>

The runtime argument for the kernel.

fn compilation_arg<'a, R: Runtime>( runtime_arg: &Self::RuntimeArg<'a, R>, ) -> Self::CompilationArg

impl LaunchArgExpand for FastDivmod

type CompilationArg = FastDivmodCompilationArg

Compilation argument.

fn expand( arg: &Self::CompilationArg, builder: &mut KernelBuilder, ) -> <Self as CubeType>::ExpandType

Register an input variable during compilation that fill the KernelBuilder.

fn expand_output( arg: &Self::CompilationArg, builder: &mut KernelBuilder, ) -> <Self as CubeType>::ExpandType

Register an output variable during compilation that fill the KernelBuilder.

impl Copy for FastDivmod