|
MayaFlux 1.0.0
Digital-First Multimedia Processing Framework
|
|
inlinenodiscard |
One nn_conv1d_grouped dispatch: [batch, c_in, t_in] to [batch, c_out, grouped_conv1d_length].
| layer | The layer. Both channel counts must be multiples of groups. |
| src | Device address of the input. |
| weights | Device address of the weights, [c_out, c_in / groups, kernel]. |
| bias | Device address of the bias, [c_out], or 0 for none. |
| dst | Device address of the output. |
| batch | Independent repeats of the input. |
| t_in | Input length. |
| std::invalid_argument | When the layer is empty, the groups do not divide the channels or the output would be empty. |
Definition at line 110 of file Stages.hpp.
References MayaFlux::Yantra::TensorKernels::GroupedConv1d::activation, MayaFlux::Yantra::TensorKernels::GroupedConv1d::c_in, MayaFlux::Yantra::TensorKernels::GroupedConv1d::c_out, conv1d_activation_shift, conv1d_flag_bias, grouped_conv1d_length(), grouped_conv1d_shader, MayaFlux::Yantra::TensorKernels::GroupedConv1d::groups, MayaFlux::Yantra::TensorKernels::GroupedConv1d::kernel, MayaFlux::Yantra::TensorKernels::detail::make_grid_stage(), MayaFlux::Yantra::TensorKernels::GroupedConv1d::pad, MayaFlux::Yantra::TensorKernels::detail::refuse(), and MayaFlux::Yantra::TensorKernels::GroupedConv1d::stride.
Referenced by MayaFlux::Yantra::HubertPhonemeModel::Impl::run().