A high-performance inference engine for AI models
// Autogenerated stub
import definitions;
[[Variants("InputT", "float, half, BFloat16")]]
[[Variants("AffineT", "float, half, BFloat16")]]
[[Variants("OutputT", "float, half, BFloat16")]]
[[Variants("AccumT", "float")]]
[[Public]] [[Kernel]] func Normalization<InputT, AffineT, OutputT, AccumT>(
[[Optional("!in_place")]] StructuredBuffer<InputT> input,
[[Optional("has_scales")]] StructuredBuffer<AffineT> scales,
[[Optional("has_biases")]] StructuredBuffer<AffineT> biases,
RWStructuredBuffer<OutputT> output,
[[Optional("copy_to_shortcut")]] RWStructuredBuffer<InputT> shortcut,
[[Optional("use_hadamard")]] StructuredBuffer<int32_t> hadamard_factors,
uint32_t batch_size,
uint32_t element_count,
float epsilon,
float scale_offset,
float post_layer_scalar,
[[Specialize]] bool in_place,
[[Specialize]] bool subtract_mean,
[[Specialize]] bool full_layer,
[[Specialize]] bool copy_to_shortcut,
[[Specialize]] bool residual_add,
[[Specialize]] bool use_hadamard,
[[Specialize]] bool scale_residual_sum,
[[Specialize]] bool scale_output,
[[Specialize]] bool has_biases,
[[Specialize]] bool has_scales,
ThreadContext thread_context,
[[Groups("batch_size")]] uint32_t batch_idx,
[[Threads("1024")]] uint32_t thread_in_row,
) {
// TODO
}