Enqueues a fused convolution operation onto the stream. We provide several variants with different types for inputs, biases and scaling parameters. Arguments (all borrowed): stream: borrowed pointer to the stream that the 'convolve' operation should be enqueued onto. conv_input_descriptor: dimensions of the convolution input layer. conv_input_data: un-owned device memory region which contains the
| 1335 | // the result is the same size as the input - this requires even more |
| 1336 | // padding of the input. |
| 1337 | virtual bool DoFusedConvolve( |
| 1338 | Stream* stream, const dnn::BatchDescriptor& conv_input_descriptor, |
| 1339 | const DeviceMemory<double>& conv_input_data, double conv_input_scale, |
| 1340 | const dnn::FilterDescriptor& filter_descriptor, |
| 1341 | const DeviceMemory<double>& filter_data, |
| 1342 | const dnn::ConvolutionDescriptor& convolution_descriptor, |
| 1343 | const DeviceMemory<double>& side_input_data, double side_input_scale, |
| 1344 | const dnn::BatchDescriptor& bias_descriptor, |
| 1345 | const DeviceMemory<double>& biases, dnn::ActivationMode activation_mode, |
| 1346 | const dnn::BatchDescriptor& output_descriptor, |
| 1347 | DeviceMemory<double>* output_data, ScratchAllocator* scratch_allocator, |
| 1348 | const dnn::AlgorithmConfig& algorithm_config, |
| 1349 | dnn::ProfileResult* output_profile_result) { |
| 1350 | return false; |
| 1351 | } |
| 1352 | |
| 1353 | // This is the float version of DoFusedConvolve. |
| 1354 | virtual bool DoFusedConvolve( |
no test coverage detected