// Pointwise portion of one Elman RNN cell adjoint.
import storage;
import attributes;
struct PushConstants {
uint gates_i_index;
uint gates_h_index;
uint output_gradient_index;
uint gates_i_gradient_index;
uint gates_h_gradient_index;
uint element_count;
};
[[vk::push_constant]] PushConstants push;
[[vk::binding(0, 0)]] RWByteAddressBuffer storage_buffers[];
[kernel_name("rnn_cell_backward")]
[domain("ml")]
[variant("generic")]
[dtype("f32")]
[status("experimental")]
[shader("compute")]
[numthreads(256, 1, 1)]
void main(uint3 dispatch_thread_id : SV_DispatchThreadID) {
uint index = dispatch_thread_id.x;
if (index >= push.element_count) {
return;
}
float output = tanh(
load_f32(storage_buffers[push.gates_i_index], index)
+ load_f32(storage_buffers[push.gates_h_index], index)
);
float gradient = load_f32(storage_buffers[push.output_gradient_index], index)
* (1.0f - output * output);
store_f32(storage_buffers[push.gates_i_gradient_index], index, gradient);
store_f32(storage_buffers[push.gates_h_gradient_index], index, gradient);
}