Struct rcublas::API

pub struct API;

Defines the cuBLAS API.

Implementations

impl API

pub fn asum(
    context: &Context,
    x: *mut f32,
    result: *mut f32,
    n: i32,
    stride: Option<i32>
) -> Result<(), Error>

Compute the sum of magnitudes of the provided vector elements.

x: pointer to input vector. result: pointer to output scalar. n: number of elements to compute sum over (should not be greater than x). stride: offset from one input element to the next. Defaults to 1.

pub fn axpy(
    context: &Context,
    alpha: *mut f32,
    x: *mut f32,
    y: *mut f32,
    n: i32,
    stride_x: Option<i32>,
    stride_y: Option<i32>
) -> Result<(), Error>

Computes a vector-scalar product and adds the result to a vector.

alpha: pointer to input scalar. x: pointer to input vector. y: pointer to output vector. n: number of elements to use for operation (should not be greater than number of elements in x or y). stride_x: offset from one element in x to the next. Defaults to 1. stride_y: offset from one element in y to the next. Defaults to 1.

pub fn copy(
    context: &Context,
    x: *mut f32,
    y: *mut f32,
    n: i32,
    stride_x: Option<i32>,
    stride_y: Option<i32>
) -> Result<(), Error>

Copies a vector into another vector.

x: pointer to input vector. y: pointer to output vector. n: number of elements to use for operation (should not be greater than number of elements in x or y). stride_x: offset from one element in x to the next. Defaults to 1. stride_y: offset from one element in y to the next. Defaults to 1.

pub fn dot(
    context: &Context,
    x: *mut f32,
    y: *mut f32,
    result: *mut f32,
    n: i32,
    stride_x: Option<i32>,
    stride_y: Option<i32>
) -> Result<(), Error>

TODO: DOC

pub fn nrm2(
    context: &Context,
    x: *mut f32,
    result: *mut f32,
    n: i32,
    stride_x: Option<i32>
) -> Result<(), Error>

TODO: DOC

pub fn scal(
    context: &Context,
    alpha: *mut f32,
    x: *mut f32,
    n: i32,
    stride_x: Option<i32>
) -> Result<(), Error>

TODO: DOC

pub fn swap(
    context: &Context,
    x: *mut f32,
    y: *mut f32,
    n: i32,
    stride_x: Option<i32>,
    stride_y: Option<i32>
) -> Result<(), Error>

TODO: DOC

impl API

pub fn gemm(
    context: &Context,
    transa: Operation,
    transb: Operation,
    m: i32,
    n: i32,
    k: i32,
    alpha: *mut f32,
    a: *mut f32,
    lda: i32,
    b: *mut f32,
    ldb: i32,
    beta: *mut f32,
    c: *mut f32,
    ldc: i32
) -> Result<(), Error>

Performs a general matrix-matrix multiplication.

Note: the matrices are expected to be ordered column-major (FORTRAN-style).

impl API

pub fn create() -> Result<Context, Error>

Create a new cuBLAS context, allocating resources on the host and the GPU.

The returned Context must be provided to future cuBLAS calls. Creating contexts all the time can lead to performance problems. Generally one Context per GPU device and configuration is recommended.

pub unsafe fn destroy(context: &mut Context) -> Result<(), Error>

Destroys the cuBLAS context, freeing its resources.

Should generally not be called directly. Automatically called when dropping a Context.

Safety

Instructs CUDA to remove the cuBLAS handle, causing any further instructions to fail. This should be called at the end of using cuBLAS and should ideally be handled by drop exclusively, and never called by the user.

pub fn get_version(context: &Context) -> Result<i32, Error>

Get CUBLAS Version

pub fn get_pointer_mode(context: &Context) -> Result<PointerMode, Error>

Retrieve the pointer mode for a given cuBLAS context.

pub fn set_pointer_mode(
    context: &mut Context,
    pointer_mode: PointerMode
) -> Result<(), Error>

Set the pointer mode for a given cuBLAS context.

Trait Implementations

impl Clone for API

fn clone(&self) -> API

Returns a copy of the value.

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

Performs copy-assignment from source.

impl Debug for API

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

Formats the value using the given formatter.

impl Copy for API