Trait collenchyma_nn::Convolution [−] [src]

pub trait Convolution<F>: NN<F> {
    fn new_convolution_config(&self, src: &SharedTensor<F>, dest: &SharedTensor<F>, filter: &mut SharedTensor<F>, algo_fwd: ConvForwardAlgo, algo_bwd_filter: ConvBackwardFilterAlgo, algo_bwd_data: ConvBackwardDataAlgo, stride: &[i32], zero_padding: &[i32]) -> Result<Self::CC, Error>;
    fn convolution(&self, filter: &mut SharedTensor<F>, x: &mut SharedTensor<F>, result: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>;
    fn convolution_plain(&self, filter: &SharedTensor<F>, x: &SharedTensor<F>, result: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>;
    fn convolution_grad_filter(&self, src_data: &mut SharedTensor<F>, dest_diff: &mut SharedTensor<F>, filter_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>;
    fn convolution_grad_filter_plain(&self, src_data: &SharedTensor<F>, dest_diff: &SharedTensor<F>, filter_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>;
    fn convolution_grad_data(&self, filter: &mut SharedTensor<F>, x_diff: &mut SharedTensor<F>, result_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>;
    fn convolution_grad_data_plain(&self, filter: &SharedTensor<F>, x_diff: &SharedTensor<F>, result_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>;
}

Provides the functionality for a Backend to support Convolution operations.

Required Methods

fn new_convolution_config(&self, src: &SharedTensor<F>, dest: &SharedTensor<F>, filter: &mut SharedTensor<F>, algo_fwd: ConvForwardAlgo, algo_bwd_filter: ConvBackwardFilterAlgo, algo_bwd_data: ConvBackwardDataAlgo, stride: &[i32], zero_padding: &[i32]) -> Result<Self::CC, Error>

Creates a new ConvolutionConfig, which needs to be passed to further convolution Operations.

fn convolution(&self, filter: &mut SharedTensor<F>, x: &mut SharedTensor<F>, result: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>

Computes a CNN convolution over the input Tensor x with complete memory management.

Saves the result to result.

For a no-memory managed version see convolution_plain.

fn convolution_plain(&self, filter: &SharedTensor<F>, x: &SharedTensor<F>, result: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>

Computes the convolution over the input Tensor x without any memory management.

Saves the result to result.

Attention:
For a correct computation result, you need to manage the memory allocation and synchronization yourself.
For a memory managed version see convolution.

fn convolution_grad_filter(&self, src_data: &mut SharedTensor<F>, dest_diff: &mut SharedTensor<F>, filter_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>

Computes the gradient of a CNN convolution with respect to the filter and complete memory management.

Saves the result to filter_diff.

For a no-memory managed version see convolution_grad_filter_plain.

fn convolution_grad_filter_plain(&self, src_data: &SharedTensor<F>, dest_diff: &SharedTensor<F>, filter_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>

Computes the gradient of a convolution with respect to the filter and without any memory management.

Saves the result to filter_diff.

Attention:
For a correct computation result, you need to manage the memory allocation and synchronization yourself.
For a memory managed version see convolution_grad_filter.

fn convolution_grad_data(&self, filter: &mut SharedTensor<F>, x_diff: &mut SharedTensor<F>, result_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>

Computes the gradient of a CNN convolution over the input Tensor x with respect to the data and complete memory management.

Saves the result to result_diff.

For a no-memory managed version see convolution_grad_data_plain.

fn convolution_grad_data_plain(&self, filter: &SharedTensor<F>, x_diff: &SharedTensor<F>, result_diff: &mut SharedTensor<F>, workspace: &mut SharedTensor<u8>, config: &Self::CC) -> Result<(), Error>

Computes the gradient of a convolution over the input Tensor x with respect to the data and without any memory management.

Saves the result to result_diff.

Attention:
For a correct computation result, you need to manage the memory allocation and synchronization yourself.
For a memory managed version see convolution_grad_data.

Implementors