Struct CudaBackend

pub struct CudaBackend { /* private fields */ }

A structure of CUDA backend.

Implementations

impl CudaBackend

pub fn new() -> Result<CudaBackend>

Creates a CUDA backend for a first device.

pub fn new_with_ordinal_and_flags( ordinal: usize, is_cublas: bool, is_mma: bool, ) -> Result<CudaBackend>

Creates a CUDA backend with the ordinal number and the flags.

This method takes the following flags:

• is_cublas - use the cuBLAS library to multiplication of matrices
• is_mma - use the mma instruction to multiplication of matrices

pub fn has_cublas(&self) -> bool

Trait Implementations

impl Backend for CudaBackend

fn name(&self) -> &'static str

Returns the backend name.

fn has_cublas(&self) -> bool

Returns true if the backend uses cuBLAS, otherwise false.

unsafe fn alloc(&self, n: usize) -> Result<BackendArray>

Allocates a backend array.

fn alloc_and_store_zeros(&self, n: usize) -> Result<BackendArray>

Allocates a backend array and stores zeros in the backend array.

fn alloc_and_store(&self, elems: &[f32]) -> Result<BackendArray>

Allocates a backend array and stores the elements in the backend array.

fn load(&self, a: &BackendArray, elems: &mut [f32]) -> Result<()>

Loads elements from the backenc array.

fn store(&self, a: &BackendArray, elems: &[f32]) -> Result<()>

Stores elements in the backend array.

fn copy(&self, a: &BackendArray, b: &BackendArray) -> Result<()>

Copies the a backend array to the b backend array.

fn transpose_a( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Transposes the a matrix and then the result is in the b matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}$).

fn add_a_b( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Adds the a matrix onto the b matrix and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}+\mathbf{B}$).

fn add_at_b( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Adds the transposed a matrix onto the b matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}+\mathbf{B}$).

fn add_a_bt( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Adds the a matrix onto the transposed b matrix and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}+{\mathbf{B}}^{\mathrm{T}}$).

fn add_at_bt( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Adds the transposed a matrix onto the transposed b matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}+{\mathbf{B}}^{\mathrm{T}}$).

fn sub_a_b( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the b matrix from the a matrix and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}-\mathbf{B}$).

fn sub_at_b( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the b matrix from the transposed a matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}-\mathbf{B}$).

fn sub_a_bt( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the transposed b matrix from the a matrix and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}-{\mathbf{B}}^{\mathrm{T}}$).

fn sub_at_bt( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the transposed b matrix from the transposed a matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}-{\mathbf{B}}^{\mathrm{T}}$).

fn mul_a_b( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, l: usize, ) -> Result<()>

Multiplies the a matrix by the b matrix and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}·\mathbf{B}$).

fn mul_at_b( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, l: usize, ) -> Result<()>

Multiplies the transposed a matrix by the b matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}·\mathbf{B}$).

fn mul_a_bt( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, l: usize, ) -> Result<()>

Multiplies the a matrix by the transposed b matrix and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}·{\mathbf{B}}^{\mathrm{T}}$).

fn mul_at_bt( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, l: usize, ) -> Result<()>

Multiplies the transposed a matrix by the transposed b matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}·{\mathbf{B}}^{\mathrm{T}}$).

fn mul_a_b_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Multiplies the a matrix elements by the b matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=a_{\mathit{ij}}·b_{\mathit{ij}}$).

fn mul_at_b_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Multiplies the transposed a matrix elements by the b matrix elements and saves the result to the c matrix ($c_{\mathit{ij}}=a_{\mathit{ji}}·b_{\mathit{ij}}$).

fn mul_a_bt_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Multiplies the a matrix elements by the transposed b matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=a_{\mathit{ij}}·b_{\mathit{ji}}$).

fn mul_at_bt_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Multiplies the transposed a matrix elements by the transposed b matrix elements and then the result is in the c matrix. ($c_{\mathit{ij}}=a_{\mathit{ji}}·b_{\mathit{ji}}$).

fn div_a_b_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the a matrix elements by the b matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=\frac{a_{\mathit{ij}}}{b_{\mathit{ij}}}$).

fn div_at_b_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the transposed a matrix elements by the b matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=\frac{a_{\mathit{ji}}}{b_{\mathit{ij}}}$).

fn div_a_bt_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the transposed a matrix elements by the b matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=\frac{a_{\mathit{ij}}}{b_{\mathit{ji}}}$).

fn div_at_bt_for_elems( &self, a: &BackendArray, b: &BackendArray, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the transposed a matrix elements by the transposed b matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=\frac{a_{\mathit{ji}}}{b_{\mathit{ji}}}$).

fn add_a_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Adds the a matrix onto the b scalar and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}+b$).

fn add_at_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Adds the transposed a matrix onto the b scalar and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}+b$).

fn sub_a_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the b scalar from the a matrix and then the result is in the c matrix. ($\mathbf{C}=\mathbf{A}-b$).

fn sub_at_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the b scalar from the transposed a matrix and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}-b$).

fn rsub_a_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the a matrix from the b scalar and then the result is in the c matrix ($\mathbf{C}=b-\mathbf{A}$).

fn rsub_at_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Subtracts the transposed a matrix from the b scalar and then the result is in the c matrix ($\mathbf{C}=b-{\mathbf{A}}^{\mathrm{T}}$).

fn mul_a_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Multiplies the a matrix by the b scalar and then the result is in the c matrix ($\mathbf{C}=\mathbf{A}·b$).

fn mul_at_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Multiplies the transposed a matrix by the b scalar and then the result is in the c matrix ($\mathbf{C}={\mathbf{A}}^{\mathrm{T}}·b$).

fn div_a_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the a matrix by the b scalar and then the result is in the c matrix ($\mathbf{C}=\frac{\mathbf{A}}{b}$).

fn div_at_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the transposed a matrix by the b scalar and then the result is in the c matrix ($\mathbf{C}=\frac{{\mathbf{A}}^{\mathrm{T}}}{b}$).

fn rdiv_a_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the b scalar by the a matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=\frac{b}{a_{\mathit{ij}}}$).

fn rdiv_at_b_for_scalar( &self, a: &BackendArray, b: f32, c: &BackendArray, n: usize, m: usize, ) -> Result<()>

Divides the b scalar by the transposed a matrix elements and then the result is in the c matrix ($c_{\mathit{ij}}=\frac{b}{a_{\mathit{ji}}}$).

fn sigmoid_a( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Calculates sigmoid function for the a matrix adn the result is the b matrix ($\mathbf{B}=\mathrm{sigmoid}\left(\mathbf{A}\right)$).

fn sigmoid_at( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Calculates sigmoid function for the transposed a matrix and then the result is in the b matrix ($\mathbf{B}=\mathrm{sigmoid}\left({\mathbf{A}}^{\mathrm{T}}\right)$).

fn tanh_a( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Calculates hyperbolic tangent function for the a matrix and then the result is in b matrix ($\mathbf{B}=\tanh \left(\mathbf{A}\right)$).

fn tanh_at( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Calculates hyperbolic tangent function for the transposed a matrix and then the result is in the b matrix ($\mathbf{B}=\tanh \left({\mathbf{A}}^{\mathrm{T}}\right)$).

fn softmax_a( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Calculates softmax function for the a matrix and then the result is in the b matrix ($\mathbf{B}=\mathrm{softmax}\left(\mathbf{A}\right)$).

fn softmax_at( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Calculates softmax function for the transposed a matrix and then the result is in the b matrix ($\mathbf{B}=\mathrm{softmax}\left({\mathbf{A}}^{\mathrm{T}}\right)$).

fn repeat_col_a( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Repeats the a vector as column ($b_{\mathit{ij}}=a_i$).

fn repeat_row_a( &self, a: &BackendArray, b: &BackendArray, n: usize, m: usize, ) -> Result<()>

Repeats the a vector as row ($b_{\mathit{ij}}=a_j$).