#include <cudnn_frontend.h>
#include "opennn/core/log.h"
#include <algorithm>
#include <filesystem>
#include <fstream>
#include <string_view>
#include "opennn/core/device_backend.h"
#include "opennn/core/cuda/energy_meter.h"
#include "opennn/core/profiler.h"
#include "opennn/core/string_utilities.h"
#include "opennn/core/tensor_types.h"

Classes
struct	opennn::cudnn_frontend::GraphSlot

Namespaces
namespace	opennn

namespace	opennn::cudnn_frontend

Typedefs
using	opennn::cudnn_frontend::VariantPack = unordered_map<shared_ptr<graph::Tensor_attributes>, void*>

Functions
int	opennn::cudnn_frontend::device_sm_version ()

bool	opennn::cudnn_frontend::frontend_enabled ()

bool	opennn::cudnn_frontend::frontend_verbose ()

bool	opennn::cudnn_frontend::bn_frontend_enabled ()

bool	opennn::cudnn_frontend::deterministic_algorithms_enabled ()

bool	opennn::cudnn_frontend::conv_energy_autotune_enabled ()

float	opennn::cudnn_frontend::conv_energy_tolerance ()

int64_t	opennn::cudnn_frontend::conv_energy_window_ms ()

bool	opennn::cudnn_frontend::conv_energy_verbose ()

bool	opennn::cudnn_frontend::conv_energy_stage_active ()

bool	opennn::cudnn_frontend::graph_timing_enabled ()

profiler::Stats &	opennn::cudnn_frontend::graph_timing_stats ()

template<typename TensorMap>
void	opennn::cudnn_frontend::execute_graph (graph::Graph &graph, TensorMap &tensors, void *workspace, const string &what, const string &timing_label)

void *	opennn::cudnn_frontend::shared_workspace (int64_t bytes)

template<typename GraphCache, typename Body>
bool	opennn::cudnn_frontend::run_frontend (GraphCache &cache, const char *label, Body &&body)

void	opennn::cudnn_frontend::throw_frontend_unavailable (const string &what)

int64_t	opennn::cudnn_frontend::element_bytes (DataType_t dtype)

DataType_t	opennn::cudnn_frontend::to_dtype (Type t)

vector< int64_t >	opennn::cudnn_frontend::nhwc_strides (int64_t c, int64_t h, int64_t w)

shared_ptr< graph::Graph >	opennn::cudnn_frontend::new_graph (Type dtype=Type::FP32)

shared_ptr< graph::Tensor_attributes >	opennn::cudnn_frontend::nhwc_tensor (graph::Graph &graph, const char *name, int64_t n, int64_t c, int64_t h, int64_t w)

void	opennn::cudnn_frontend::set_nhwc_output (shared_ptr< graph::Tensor_attributes > &tensor, int64_t n, int64_t c, int64_t h, int64_t w)

vector< int64_t >	opennn::cudnn_frontend::bhsd_strides (int64_t h, int64_t s, int64_t d, bool interleaved)

shared_ptr< graph::Tensor_attributes >	opennn::cudnn_frontend::bhsd_tensor (graph::Graph &graph, const char *name, int64_t b, int64_t h, int64_t s, int64_t d, bool interleaved)

void	opennn::cudnn_frontend::set_bhsd_output (shared_ptr< graph::Tensor_attributes > &tensor, int64_t b, int64_t h, int64_t s, int64_t d, bool interleaved)

shared_ptr< graph::Tensor_attributes >	opennn::cudnn_frontend::per_channel_tensor (graph::Graph &graph, const char *name, int64_t channels)

void	opennn::cudnn_frontend::set_per_channel_output (shared_ptr< graph::Tensor_attributes > &tensor, int64_t channels)

template<typename... Args>
string	opennn::cudnn_frontend::timing_label (format_string< Args... > fmt, Args &&... args)

shared_ptr< graph::Tensor_attributes >	opennn::cudnn_frontend::scalar_tensor (graph::Graph &graph, const char *name, DataType_t dtype, bool pass_by_value=false, int64_t batch=1)

int64_t	opennn::cudnn_frontend::candidate_limit_from (long long value)

int64_t	opennn::cudnn_frontend::autotune_candidate_limit ()

int64_t	opennn::cudnn_frontend::autotune_candidate_limit (const string &tag)

vector< HeurMode_t >	opennn::cudnn_frontend::heuristic_modes ()

const vector< NumericalNote_t > &	opennn::cudnn_frontend::conv_engine_notes ()

bool	opennn::cudnn_frontend::build_top_candidates (graph::Graph &graph, int64_t limit)

bool	opennn::cudnn_frontend::plan_cache_enabled ()

bool	opennn::cudnn_frontend::sdpa_autotune_enabled ()

int64_t	opennn::cudnn_frontend::sdpa_workspace_cap_override_bytes ()

const std::filesystem::path &	opennn::cudnn_frontend::plan_cache_directory ()

std::filesystem::path	opennn::cudnn_frontend::plan_cache_file (const graph::Graph &graph)

bool	opennn::cudnn_frontend::load_cached_plan (graph::Graph &graph, const cudnnHandle_t handle, int64_t &workspace_bytes)

void	opennn::cudnn_frontend::store_cached_plan (graph::Graph &graph)

int64_t	opennn::cudnn_frontend::autotune_workspace_bytes (const graph::Graph &graph)

int64_t	opennn::cudnn_frontend::sdpa_workspace_cap_bytes (int64_t batch, int64_t heads, int64_t q_seq, int64_t src_seq, int64_t head_dim)

bool	opennn::cudnn_frontend::finalize_attention (graph::Graph &graph, const string &tag, int64_t &workspace_bytes, bool allow_autotune=false, int64_t workspace_cap=0)

shared_ptr< graph::Tensor_attributes >	opennn::cudnn_frontend::seq_len_scalar (graph::Graph &graph, const char *name, int64_t batch=1)

bool	opennn::cudnn_frontend::finalize (graph::Graph &graph, int64_t &workspace_bytes, const string &tag, const bool fp32_graph=false)

void	opennn::cudnn_frontend::report_autotune_skipped (const char tag, const char reason)

bool	opennn::cudnn_frontend::is_convolution_tag (const char *tag)

template<typename TensorMap>
void	opennn::cudnn_frontend::select_plan_by_energy (graph::Graph &graph, TensorMap &tensors, void workspace, const char tag)

template<typename TensorMap>
void	opennn::cudnn_frontend::autotune_now (bool &pending, graph::Graph &graph, TensorMap &tensors, int64_t &workspace_bytes, const char *tag=nullptr)

template<typename TensorMap>
void	opennn::cudnn_frontend::autotune_with_scratch (bool &pending, graph::Graph &graph, const TensorMap &tensors, int64_t &workspace_bytes, const char *tag=nullptr)

template<typename TensorMap>
void	opennn::cudnn_frontend::run_slot (GraphSlot &slot, TensorMap &tensors, const char *what, const string &timing_label, bool scratch_autotune)

Variables
constexpr size_t	opennn::cudnn_frontend::graph_cache_capacity = 8

const auto	opennn::cudnn_frontend::check_status

Classes

Namespaces

Typedefs

Functions

Variables