version:2.1.5
bugfixes: update:拨号使用ocr进行识别
This commit is contained in:
@@ -19,7 +19,6 @@
|
||||
|
||||
#ifndef PADDLE_LITE_API_H_ // NOLINT
|
||||
#define PADDLE_LITE_API_H_
|
||||
|
||||
#include <map>
|
||||
#include <memory>
|
||||
#include <string>
|
||||
@@ -28,567 +27,488 @@
|
||||
#include "paddle_place.h" // NOLINT
|
||||
|
||||
namespace paddle {
|
||||
namespace lite_api {
|
||||
namespace lite_api {
|
||||
|
||||
using shape_t = std::vector<int64_t>;
|
||||
using lod_t = std::vector<std::vector<uint64_t>>;
|
||||
using shape_t = std::vector<int64_t>;
|
||||
using lod_t = std::vector<std::vector<uint64_t>>;
|
||||
|
||||
enum class LiteModelType {
|
||||
kProtobuf = 0, kNaiveBuffer, UNK
|
||||
};
|
||||
enum class LiteModelType { kProtobuf = 0, kNaiveBuffer, UNK };
|
||||
// Methods for allocating L3Cache on Arm platform
|
||||
enum class L3CacheSetMethod {
|
||||
kDeviceL3Cache = 0, // Use the system L3 Cache size, best performance.
|
||||
kDeviceL2Cache = 1, // Use the system L2 Cache size, trade off performance
|
||||
// with less memory consumption.
|
||||
kAbsolute = 2, // Use the external setting.
|
||||
// kAutoGrow = 3, // Not supported yet, least memory consumption.
|
||||
};
|
||||
enum class L3CacheSetMethod {
|
||||
kDeviceL3Cache = 0, // Use the system L3 Cache size, best performance.
|
||||
kDeviceL2Cache = 1, // Use the system L2 Cache size, trade off performance
|
||||
// with less memory consumption.
|
||||
kAbsolute = 2, // Use the external setting.
|
||||
// kAutoGrow = 3, // Not supported yet, least memory consumption.
|
||||
};
|
||||
|
||||
// return true if current device supports OpenCL model
|
||||
LITE_API bool IsOpenCLBackendValid(bool check_fp16_valid = false);
|
||||
LITE_API bool IsOpenCLBackendValid(bool check_fp16_valid = false);
|
||||
|
||||
// return current opencl device type,
|
||||
// if opencl not enabled or IsOpenCLBackendValid return false, it will return -1
|
||||
// UNKNOWN:0, QUALCOMM_ADRENO:1, ARM_MALI:2, IMAGINATION_POWERVR:3, OTHERS:4,
|
||||
LITE_API int GetOpenCLDeviceType();
|
||||
LITE_API int GetOpenCLDeviceType();
|
||||
|
||||
struct LITE_API Tensor {
|
||||
explicit Tensor(void *raw);
|
||||
struct LITE_API Tensor {
|
||||
explicit Tensor(void* raw);
|
||||
explicit Tensor(const void* raw);
|
||||
|
||||
explicit Tensor(const void *raw);
|
||||
void Resize(const shape_t& shape);
|
||||
|
||||
void Resize(const shape_t &shape);
|
||||
/// Readonly data.
|
||||
template <typename T>
|
||||
const T* data() const;
|
||||
|
||||
/// Readonly data.
|
||||
template<typename T>
|
||||
const T *data() const;
|
||||
template <typename T>
|
||||
T* mutable_data(TargetType type = TargetType::kHost) const;
|
||||
|
||||
template<typename T>
|
||||
T *mutable_data(TargetType type = TargetType::kHost) const;
|
||||
void* mutable_metal_data(void* ptr) const;
|
||||
|
||||
void *mutable_metal_data(void *ptr) const;
|
||||
// Share external memory. Note: ensure that the data pointer is in a valid
|
||||
// state
|
||||
// during the prediction process.
|
||||
void ShareExternalMemory(void* data, size_t memory_size, TargetType target);
|
||||
|
||||
// Share external memory. Note: ensure that the data pointer is in a valid
|
||||
// state
|
||||
// during the prediction process.
|
||||
void ShareExternalMemory(void *data, size_t memory_size, TargetType target);
|
||||
template <typename T, TargetType type = TargetType::kHost>
|
||||
void CopyFromCpu(const T* data);
|
||||
|
||||
template<typename T, TargetType type = TargetType::kHost>
|
||||
void CopyFromCpu(const T *data);
|
||||
template <typename T>
|
||||
void CopyToCpu(T* data) const;
|
||||
/// Shape of the tensor.
|
||||
shape_t shape() const;
|
||||
TargetType target() const;
|
||||
PrecisionType precision() const;
|
||||
void SetPrecision(PrecisionType precision);
|
||||
|
||||
template<typename T>
|
||||
void CopyToCpu(T *data) const;
|
||||
// LoD of the tensor
|
||||
lod_t lod() const;
|
||||
|
||||
/// Shape of the tensor.
|
||||
shape_t shape() const;
|
||||
// Set LoD of the tensor
|
||||
void SetLoD(const lod_t& lod);
|
||||
bool IsInitialized() const;
|
||||
|
||||
TargetType target() const;
|
||||
|
||||
PrecisionType precision() const;
|
||||
|
||||
void SetPrecision(PrecisionType precision);
|
||||
|
||||
// LoD of the tensor
|
||||
lod_t lod() const;
|
||||
|
||||
// Set LoD of the tensor
|
||||
void SetLoD(const lod_t &lod);
|
||||
|
||||
bool IsInitialized() const;
|
||||
|
||||
private:
|
||||
void *raw_tensor_;
|
||||
};
|
||||
private:
|
||||
void* raw_tensor_;
|
||||
};
|
||||
|
||||
/// The PaddlePredictor defines the basic interfaces for different kinds of
|
||||
/// predictors.
|
||||
class LITE_API PaddlePredictor {
|
||||
public:
|
||||
PaddlePredictor() = default;
|
||||
class LITE_API PaddlePredictor {
|
||||
public:
|
||||
PaddlePredictor() = default;
|
||||
|
||||
/// Get i-th input.
|
||||
virtual std::unique_ptr<Tensor> GetInput(int i) = 0;
|
||||
/// Get i-th input.
|
||||
virtual std::unique_ptr<Tensor> GetInput(int i) = 0;
|
||||
|
||||
/// Get i-th output.
|
||||
virtual std::unique_ptr<const Tensor> GetOutput(int i) const = 0;
|
||||
/// Get i-th output.
|
||||
virtual std::unique_ptr<const Tensor> GetOutput(int i) const = 0;
|
||||
|
||||
virtual void Run() = 0;
|
||||
virtual void Run() = 0;
|
||||
virtual std::shared_ptr<PaddlePredictor> Clone() = 0;
|
||||
virtual std::shared_ptr<PaddlePredictor> Clone(
|
||||
const std::vector<std::string>& var_names) = 0;
|
||||
|
||||
virtual std::shared_ptr<PaddlePredictor> Clone() = 0;
|
||||
virtual std::string GetVersion() const = 0;
|
||||
|
||||
virtual std::shared_ptr<PaddlePredictor> Clone(
|
||||
const std::vector<std::string> &var_names) = 0;
|
||||
// Get input names
|
||||
virtual std::vector<std::string> GetInputNames() = 0;
|
||||
// Get output names
|
||||
virtual std::vector<std::string> GetOutputNames() = 0;
|
||||
// Get output names
|
||||
virtual std::vector<std::string> GetParamNames();
|
||||
|
||||
virtual std::string GetVersion() const = 0;
|
||||
/// Release all tmp tensor to compress the size of the memory pool.
|
||||
virtual bool TryShrinkMemory() = 0;
|
||||
|
||||
// Get input names
|
||||
virtual std::vector<std::string> GetInputNames() = 0;
|
||||
// Get Input by name
|
||||
virtual std::unique_ptr<Tensor> GetInputByName(const std::string& name) = 0;
|
||||
|
||||
// Get output names
|
||||
virtual std::vector<std::string> GetOutputNames() = 0;
|
||||
/// Get a readonly tensor, return null if no one called `name` exists.
|
||||
virtual std::unique_ptr<const Tensor> GetTensor(
|
||||
const std::string& name) const = 0;
|
||||
/// Get a mutable tensor, return null if on one called `name` exists
|
||||
/// internal infereces API, not recommanded.
|
||||
virtual std::unique_ptr<Tensor> GetMutableTensor(const std::string& name);
|
||||
|
||||
// Get output names
|
||||
virtual std::vector<std::string> GetParamNames();
|
||||
/// Persist the optimized model to disk. This API is only supported by
|
||||
/// CxxConfig, and the persisted model can be reused for MobileConfig.
|
||||
virtual void SaveOptimizedModel(
|
||||
const std::string& model_dir,
|
||||
LiteModelType model_type = LiteModelType::kProtobuf,
|
||||
bool record_info = false);
|
||||
|
||||
/// Release all tmp tensor to compress the size of the memory pool.
|
||||
virtual bool TryShrinkMemory() = 0;
|
||||
virtual ~PaddlePredictor() = default;
|
||||
|
||||
// Get Input by name
|
||||
virtual std::unique_ptr<Tensor> GetInputByName(const std::string &name) = 0;
|
||||
|
||||
/// Get a readonly tensor, return null if no one called `name` exists.
|
||||
virtual std::unique_ptr<const Tensor> GetTensor(
|
||||
const std::string &name) const = 0;
|
||||
|
||||
/// Get a mutable tensor, return null if on one called `name` exists
|
||||
/// internal infereces API, not recommanded.
|
||||
virtual std::unique_ptr<Tensor> GetMutableTensor(const std::string &name);
|
||||
|
||||
/// Persist the optimized model to disk. This API is only supported by
|
||||
/// CxxConfig, and the persisted model can be reused for MobileConfig.
|
||||
virtual void SaveOptimizedModel(
|
||||
const std::string &model_dir,
|
||||
LiteModelType model_type = LiteModelType::kProtobuf,
|
||||
bool record_info = false);
|
||||
|
||||
virtual ~PaddlePredictor() = default;
|
||||
|
||||
protected:
|
||||
int threads_{1};
|
||||
lite_api::PowerMode mode_{lite_api::LITE_POWER_NO_BIND};
|
||||
};
|
||||
protected:
|
||||
int threads_{1};
|
||||
lite_api::PowerMode mode_{lite_api::LITE_POWER_NO_BIND};
|
||||
};
|
||||
|
||||
/// Base class for all the configs.
|
||||
class LITE_API ConfigBase {
|
||||
std::string model_dir_;
|
||||
int threads_{1};
|
||||
PowerMode mode_{LITE_POWER_NO_BIND};
|
||||
// gpu opencl
|
||||
CLTuneMode opencl_tune_mode_{CL_TUNE_NONE};
|
||||
std::string opencl_bin_path_{""};
|
||||
std::string opencl_bin_name_{""};
|
||||
CLPrecisionType opencl_precision_{CL_PRECISION_AUTO};
|
||||
// Where to cache the npu/xpu/rknpu/apu offline model to the binary files
|
||||
std::string subgraph_model_cache_dir_{""};
|
||||
// Set the cached npu/xpu/rknpu/apu offline model from the buffers
|
||||
std::map<std::string, std::pair<std::vector<char>, std::vector<char>>>
|
||||
subgraph_model_cache_buffers_{};
|
||||
// The selected NNAdapter devices to build and run the model.
|
||||
std::vector<std::string> nnadapter_device_names_{};
|
||||
// The NNAdapter context properties for device configuration, model
|
||||
// compilation and execution
|
||||
std::string nnadapter_context_properties_{};
|
||||
// The directory to find and store the compiled NNAdapter models.
|
||||
std::string nnadapter_model_cache_dir_{""};
|
||||
// The buffers for loading the compiled NNAdapter models from memory.
|
||||
std::map<std::string, std::vector<char>> nnadapter_model_cache_buffers_{};
|
||||
// The custom configuration file or buffer for the NNAdapter subgraph
|
||||
// partition, here is an example:
|
||||
// op_type:in_var_name_0,in_var_name1:out_var_name_0,out_var_name1
|
||||
// op_type::out_var_name_0
|
||||
// op_type:in_var_name_0
|
||||
// op_type
|
||||
std::string nnadapter_subgraph_partition_config_path_{};
|
||||
std::string nnadapter_subgraph_partition_config_buffer_{};
|
||||
int device_id_{0};
|
||||
int x86_math_num_threads_ = 1;
|
||||
class LITE_API ConfigBase {
|
||||
std::string model_dir_;
|
||||
int threads_{1};
|
||||
PowerMode mode_{LITE_POWER_NO_BIND};
|
||||
// gpu opencl
|
||||
CLTuneMode opencl_tune_mode_{CL_TUNE_NONE};
|
||||
std::string opencl_bin_path_{""};
|
||||
std::string opencl_bin_name_{""};
|
||||
CLPrecisionType opencl_precision_{CL_PRECISION_AUTO};
|
||||
// Where to cache the npu/xpu/rknpu/apu offline model to the binary files
|
||||
std::string subgraph_model_cache_dir_{""};
|
||||
// Set the cached npu/xpu/rknpu/apu offline model from the buffers
|
||||
std::map<std::string, std::pair<std::vector<char>, std::vector<char>>>
|
||||
subgraph_model_cache_buffers_{};
|
||||
// The selected NNAdapter devices to build and run the model.
|
||||
std::vector<std::string> nnadapter_device_names_{};
|
||||
// The NNAdapter context properties for device configuration, model
|
||||
// compilation and execution
|
||||
std::string nnadapter_context_properties_{};
|
||||
// The directory to find and store the compiled NNAdapter models.
|
||||
std::string nnadapter_model_cache_dir_{""};
|
||||
// The buffers for loading the compiled NNAdapter models from memory.
|
||||
std::map<std::string, std::vector<char>> nnadapter_model_cache_buffers_{};
|
||||
// The custom configuration file or buffer for the NNAdapter subgraph
|
||||
// partition, here is an example:
|
||||
// op_type:in_var_name_0,in_var_name1:out_var_name_0,out_var_name1
|
||||
// op_type::out_var_name_0
|
||||
// op_type:in_var_name_0
|
||||
// op_type
|
||||
std::string nnadapter_subgraph_partition_config_path_{};
|
||||
std::string nnadapter_subgraph_partition_config_buffer_{};
|
||||
int device_id_{0};
|
||||
int x86_math_num_threads_ = 1;
|
||||
|
||||
std::string metal_path_;
|
||||
bool metal_use_mps_{false};
|
||||
bool metal_use_aggressive_{false};
|
||||
void *metal_device_{nullptr};
|
||||
bool metal_use_memory_reuse_{false};
|
||||
std::string metal_path_;
|
||||
bool metal_use_mps_{false};
|
||||
bool metal_use_aggressive_{false};
|
||||
void* metal_device_{nullptr};
|
||||
bool metal_use_memory_reuse_{false};
|
||||
|
||||
public:
|
||||
explicit ConfigBase(PowerMode mode = LITE_POWER_NO_BIND, int threads = 1);
|
||||
public:
|
||||
explicit ConfigBase(PowerMode mode = LITE_POWER_NO_BIND, int threads = 1);
|
||||
// set Model_dir
|
||||
void set_model_dir(const std::string& x) { model_dir_ = x; }
|
||||
const std::string& model_dir() const { return model_dir_; }
|
||||
// set Thread
|
||||
void set_threads(int threads);
|
||||
int threads() const { return threads_; }
|
||||
// set Power_mode
|
||||
void set_power_mode(PowerMode mode);
|
||||
PowerMode power_mode() const { return mode_; }
|
||||
/// \brief Set path and file name of generated OpenCL compiled kernel binary.
|
||||
///
|
||||
/// If you use GPU of specific soc, using OpenCL binary will speed up the
|
||||
/// initialization.
|
||||
///
|
||||
/// \param path Path that OpenCL compiled kernel binay file stores in. Make
|
||||
/// sure the path exist and you have Read&Write permission.
|
||||
/// \param name File name of OpenCL compiled kernel binay.
|
||||
/// \return void
|
||||
void set_opencl_binary_path_name(const std::string& path,
|
||||
const std::string& name);
|
||||
// set GPU opencl tune
|
||||
void set_opencl_tune(CLTuneMode tune_mode = CL_TUNE_NONE,
|
||||
const std::string& path = "",
|
||||
const std::string& name = "",
|
||||
size_t lws_repeats = 4);
|
||||
|
||||
// set Model_dir
|
||||
void set_model_dir(const std::string &x) { model_dir_ = x; }
|
||||
// set GPU opencl precision
|
||||
void set_opencl_precision(CLPrecisionType p = CL_PRECISION_AUTO);
|
||||
// set subgraph_model_dir
|
||||
void set_subgraph_model_cache_dir(std::string subgraph_model_cache_dir) {
|
||||
subgraph_model_cache_dir_ = subgraph_model_cache_dir;
|
||||
}
|
||||
const std::string& subgraph_model_cache_dir() const {
|
||||
return subgraph_model_cache_dir_;
|
||||
}
|
||||
void set_subgraph_model_cache_buffers(const std::string& key,
|
||||
const std::vector<char>& cfg,
|
||||
const std::vector<char>& bin);
|
||||
const std::map<std::string, std::pair<std::vector<char>, std::vector<char>>>&
|
||||
subgraph_model_cache_buffers() const {
|
||||
return subgraph_model_cache_buffers_;
|
||||
}
|
||||
// Check if the NNAdapter device is valid.
|
||||
bool check_nnadapter_device_name(const std::string& device_name);
|
||||
// Choose the NNAdapter devices to build and run the model.
|
||||
void set_nnadapter_device_names(
|
||||
const std::vector<std::string>& device_names) {
|
||||
nnadapter_device_names_ = device_names;
|
||||
}
|
||||
const std::vector<std::string>& nnadapter_device_names() const {
|
||||
return nnadapter_device_names_;
|
||||
}
|
||||
// Set the context properties by key-value map for NNAdapter device
|
||||
// configuration, model compilation and execution
|
||||
// Such as "HUAWEI_ASCEND_NPU_SELECTED_DEVICE_IDS=0;"
|
||||
void set_nnadapter_context_properties(const std::string& context_properties) {
|
||||
nnadapter_context_properties_ = context_properties;
|
||||
}
|
||||
const std::string& nnadapter_context_properties() const {
|
||||
return nnadapter_context_properties_;
|
||||
}
|
||||
// Enable caching and set the directory to search and store the compiled
|
||||
// NNAdapter models in the file system.
|
||||
void set_nnadapter_model_cache_dir(const std::string& model_cache_dir) {
|
||||
nnadapter_model_cache_dir_ = model_cache_dir;
|
||||
}
|
||||
const std::string& nnadapter_model_cache_dir() const {
|
||||
return nnadapter_model_cache_dir_;
|
||||
}
|
||||
// Set the buffers for loading the compiled NNAdapter models from memory.
|
||||
void set_nnadapter_model_cache_buffers(
|
||||
const std::string& model_cache_token,
|
||||
const std::vector<char>& model_cache_buffer);
|
||||
const std::map<std::string, std::vector<char>>&
|
||||
nnadapter_model_cache_buffers() const {
|
||||
return nnadapter_model_cache_buffers_;
|
||||
}
|
||||
// Enable the custom subgraph partition for NNAdapter by providing the
|
||||
// configuration file or buffer
|
||||
void set_nnadapter_subgraph_partition_config_path(
|
||||
const std::string& subgraph_partition_config_path) {
|
||||
nnadapter_subgraph_partition_config_path_ = subgraph_partition_config_path;
|
||||
}
|
||||
const std::string& nnadapter_subgraph_partition_config_path() const {
|
||||
return nnadapter_subgraph_partition_config_path_;
|
||||
}
|
||||
void set_nnadapter_subgraph_partition_config_buffer(
|
||||
const std::string& subgraph_partition_config_buffer) {
|
||||
nnadapter_subgraph_partition_config_buffer_ =
|
||||
subgraph_partition_config_buffer;
|
||||
}
|
||||
const std::string& nnadapter_subgraph_partition_config_buffer() const {
|
||||
return nnadapter_subgraph_partition_config_buffer_;
|
||||
}
|
||||
// set Device ID
|
||||
void set_device_id(int device_id) { device_id_ = device_id; }
|
||||
int get_device_id() const { return device_id_; }
|
||||
// set x86_math_num_threads
|
||||
void set_x86_math_num_threads(int threads);
|
||||
int x86_math_num_threads() const;
|
||||
|
||||
const std::string &model_dir() const { return model_dir_; }
|
||||
void set_metal_lib_path(const std::string& path);
|
||||
void set_metal_use_mps(bool flag);
|
||||
void set_metal_use_aggressive(bool flag);
|
||||
void set_metal_device(void* device);
|
||||
void set_metal_use_memory_reuse(bool flag);
|
||||
|
||||
// set Thread
|
||||
void set_threads(int threads);
|
||||
std::string metal_lib_path() const { return metal_path_; }
|
||||
bool metal_use_mps() const { return metal_use_mps_; }
|
||||
bool metal_use_aggressive() const { return metal_use_aggressive_; }
|
||||
void* metal_device() const { return metal_device_; }
|
||||
bool metal_use_memory_reuse() const { return metal_use_memory_reuse_; }
|
||||
};
|
||||
|
||||
int threads() const { return threads_; }
|
||||
class LITE_API CxxModelBuffer {
|
||||
public:
|
||||
CxxModelBuffer(const char* program_buffer,
|
||||
size_t program_buffer_size,
|
||||
const char* params_buffer,
|
||||
size_t params_buffer_size);
|
||||
CxxModelBuffer(std::string&& program_buffer, std::string&& params_buffer);
|
||||
const std::string& get_program() const;
|
||||
const std::string& get_params() const;
|
||||
bool is_empty() const;
|
||||
|
||||
// set Power_mode
|
||||
void set_power_mode(PowerMode mode);
|
||||
CxxModelBuffer() = default;
|
||||
CxxModelBuffer(const CxxModelBuffer&) = delete;
|
||||
|
||||
PowerMode power_mode() const { return mode_; }
|
||||
|
||||
/// \brief Set path and file name of generated OpenCL compiled kernel binary.
|
||||
///
|
||||
/// If you use GPU of specific soc, using OpenCL binary will speed up the
|
||||
/// initialization.
|
||||
///
|
||||
/// \param path Path that OpenCL compiled kernel binay file stores in. Make
|
||||
/// sure the path exist and you have Read&Write permission.
|
||||
/// \param name File name of OpenCL compiled kernel binay.
|
||||
/// \return void
|
||||
void set_opencl_binary_path_name(const std::string &path,
|
||||
const std::string &name);
|
||||
|
||||
// set GPU opencl tune
|
||||
void set_opencl_tune(CLTuneMode tune_mode = CL_TUNE_NONE,
|
||||
const std::string &path = "",
|
||||
const std::string &name = "",
|
||||
size_t lws_repeats = 4);
|
||||
|
||||
// set GPU opencl precision
|
||||
void set_opencl_precision(CLPrecisionType p = CL_PRECISION_AUTO);
|
||||
|
||||
// set subgraph_model_dir
|
||||
void set_subgraph_model_cache_dir(std::string subgraph_model_cache_dir) {
|
||||
subgraph_model_cache_dir_ = subgraph_model_cache_dir;
|
||||
}
|
||||
|
||||
const std::string &subgraph_model_cache_dir() const {
|
||||
return subgraph_model_cache_dir_;
|
||||
}
|
||||
|
||||
void set_subgraph_model_cache_buffers(const std::string &key,
|
||||
const std::vector<char> &cfg,
|
||||
const std::vector<char> &bin);
|
||||
|
||||
const std::map<std::string, std::pair<std::vector<char>, std::vector<char>>> &
|
||||
subgraph_model_cache_buffers() const {
|
||||
return subgraph_model_cache_buffers_;
|
||||
}
|
||||
|
||||
// Check if the NNAdapter device is valid.
|
||||
bool check_nnadapter_device_name(const std::string &device_name);
|
||||
|
||||
// Choose the NNAdapter devices to build and run the model.
|
||||
void set_nnadapter_device_names(
|
||||
const std::vector<std::string> &device_names) {
|
||||
nnadapter_device_names_ = device_names;
|
||||
}
|
||||
|
||||
const std::vector<std::string> &nnadapter_device_names() const {
|
||||
return nnadapter_device_names_;
|
||||
}
|
||||
|
||||
// Set the context properties by key-value map for NNAdapter device
|
||||
// configuration, model compilation and execution
|
||||
// Such as "HUAWEI_ASCEND_NPU_SELECTED_DEVICE_IDS=0;"
|
||||
void set_nnadapter_context_properties(const std::string &context_properties) {
|
||||
nnadapter_context_properties_ = context_properties;
|
||||
}
|
||||
|
||||
const std::string &nnadapter_context_properties() const {
|
||||
return nnadapter_context_properties_;
|
||||
}
|
||||
|
||||
// Enable caching and set the directory to search and store the compiled
|
||||
// NNAdapter models in the file system.
|
||||
void set_nnadapter_model_cache_dir(const std::string &model_cache_dir) {
|
||||
nnadapter_model_cache_dir_ = model_cache_dir;
|
||||
}
|
||||
|
||||
const std::string &nnadapter_model_cache_dir() const {
|
||||
return nnadapter_model_cache_dir_;
|
||||
}
|
||||
|
||||
// Set the buffers for loading the compiled NNAdapter models from memory.
|
||||
void set_nnadapter_model_cache_buffers(
|
||||
const std::string &model_cache_token,
|
||||
const std::vector<char> &model_cache_buffer);
|
||||
|
||||
const std::map<std::string, std::vector<char>> &
|
||||
nnadapter_model_cache_buffers() const {
|
||||
return nnadapter_model_cache_buffers_;
|
||||
}
|
||||
|
||||
// Enable the custom subgraph partition for NNAdapter by providing the
|
||||
// configuration file or buffer
|
||||
void set_nnadapter_subgraph_partition_config_path(
|
||||
const std::string &subgraph_partition_config_path) {
|
||||
nnadapter_subgraph_partition_config_path_ = subgraph_partition_config_path;
|
||||
}
|
||||
|
||||
const std::string &nnadapter_subgraph_partition_config_path() const {
|
||||
return nnadapter_subgraph_partition_config_path_;
|
||||
}
|
||||
|
||||
void set_nnadapter_subgraph_partition_config_buffer(
|
||||
const std::string &subgraph_partition_config_buffer) {
|
||||
nnadapter_subgraph_partition_config_buffer_ =
|
||||
subgraph_partition_config_buffer;
|
||||
}
|
||||
|
||||
const std::string &nnadapter_subgraph_partition_config_buffer() const {
|
||||
return nnadapter_subgraph_partition_config_buffer_;
|
||||
}
|
||||
|
||||
// set Device ID
|
||||
void set_device_id(int device_id) { device_id_ = device_id; }
|
||||
|
||||
int get_device_id() const { return device_id_; }
|
||||
|
||||
// set x86_math_num_threads
|
||||
void set_x86_math_num_threads(int threads);
|
||||
|
||||
int x86_math_num_threads() const;
|
||||
|
||||
void set_metal_lib_path(const std::string &path);
|
||||
|
||||
void set_metal_use_mps(bool flag);
|
||||
|
||||
void set_metal_use_aggressive(bool flag);
|
||||
|
||||
void set_metal_device(void *device);
|
||||
|
||||
void set_metal_use_memory_reuse(bool flag);
|
||||
|
||||
std::string metal_lib_path() const { return metal_path_; }
|
||||
|
||||
bool metal_use_mps() const { return metal_use_mps_; }
|
||||
|
||||
bool metal_use_aggressive() const { return metal_use_aggressive_; }
|
||||
|
||||
void *metal_device() const { return metal_device_; }
|
||||
|
||||
bool metal_use_memory_reuse() const { return metal_use_memory_reuse_; }
|
||||
};
|
||||
|
||||
class LITE_API CxxModelBuffer {
|
||||
public:
|
||||
CxxModelBuffer(const char *program_buffer,
|
||||
size_t program_buffer_size,
|
||||
const char *params_buffer,
|
||||
size_t params_buffer_size);
|
||||
|
||||
CxxModelBuffer(std::string &&program_buffer, std::string &¶ms_buffer);
|
||||
|
||||
const std::string &get_program() const;
|
||||
|
||||
const std::string &get_params() const;
|
||||
|
||||
bool is_empty() const;
|
||||
|
||||
CxxModelBuffer() = default;
|
||||
|
||||
CxxModelBuffer(const CxxModelBuffer &) = delete;
|
||||
|
||||
private:
|
||||
std::string program_;
|
||||
std::string params_;
|
||||
};
|
||||
private:
|
||||
std::string program_;
|
||||
std::string params_;
|
||||
};
|
||||
|
||||
/// CxxConfig is the config for the Full feature predictor.
|
||||
class LITE_API CxxConfig : public ConfigBase {
|
||||
std::vector<Place> valid_places_;
|
||||
std::string model_file_;
|
||||
std::string param_file_;
|
||||
std::shared_ptr<CxxModelBuffer> model_buffer_{nullptr};
|
||||
std::vector<std::string> passes_internal_{};
|
||||
bool quant_model_{false}; // Enable post_quant_dynamic in opt
|
||||
QuantType quant_type_{QuantType::QUANT_INT16};
|
||||
bool sparse_model_{false}; // Enable sparse_conv_detect_pass in opt
|
||||
float sparse_threshold_{0.6f};
|
||||
std::map<int, std::vector<std::shared_ptr<void>>>
|
||||
preferred_inputs_for_warmup_;
|
||||
class LITE_API CxxConfig : public ConfigBase {
|
||||
std::vector<Place> valid_places_;
|
||||
std::string model_file_;
|
||||
std::string param_file_;
|
||||
std::shared_ptr<CxxModelBuffer> model_buffer_{nullptr};
|
||||
std::vector<std::string> passes_internal_{};
|
||||
bool quant_model_{false}; // Enable post_quant_dynamic in opt
|
||||
QuantType quant_type_{QuantType::QUANT_INT16};
|
||||
bool sparse_model_{false}; // Enable sparse_conv_detect_pass in opt
|
||||
float sparse_threshold_{0.6f};
|
||||
std::map<int, std::vector<std::shared_ptr<void>>>
|
||||
preferred_inputs_for_warmup_;
|
||||
#ifdef LITE_WITH_CUDA
|
||||
bool multi_stream_{false};
|
||||
bool multi_stream_{false};
|
||||
#endif
|
||||
#ifdef LITE_WITH_MLU
|
||||
lite_api::MLUCoreVersion mlu_core_version_{lite_api::MLUCoreVersion::MLU_270};
|
||||
int mlu_core_number_{1};
|
||||
DataLayoutType mlu_input_layout_{DATALAYOUT(kNCHW)};
|
||||
std::vector<float> mlu_first_conv_mean_{};
|
||||
std::vector<float> mlu_first_conv_std_{};
|
||||
lite_api::MLUCoreVersion mlu_core_version_{lite_api::MLUCoreVersion::MLU_270};
|
||||
int mlu_core_number_{1};
|
||||
DataLayoutType mlu_input_layout_{DATALAYOUT(kNCHW)};
|
||||
std::vector<float> mlu_first_conv_mean_{};
|
||||
std::vector<float> mlu_first_conv_std_{};
|
||||
#endif
|
||||
|
||||
public:
|
||||
void set_valid_places(const std::vector<Place> &x) { valid_places_ = x; }
|
||||
|
||||
void set_model_file(const std::string &path) { model_file_ = path; }
|
||||
|
||||
void set_param_file(const std::string &path) { param_file_ = path; }
|
||||
|
||||
void set_model_buffer(const char *model_buffer,
|
||||
size_t model_buffer_size,
|
||||
const char *param_buffer,
|
||||
size_t param_buffer_size) {
|
||||
model_buffer_.reset(new CxxModelBuffer(
|
||||
model_buffer, model_buffer_size, param_buffer, param_buffer_size));
|
||||
}
|
||||
|
||||
void set_model_buffer(std::shared_ptr<CxxModelBuffer> model_buffer) {
|
||||
model_buffer_ = model_buffer;
|
||||
}
|
||||
|
||||
const CxxModelBuffer &get_model_buffer() const;
|
||||
|
||||
// internal inference to choose passes for model optimizing,
|
||||
// it's designed for internal developer and not recommanded
|
||||
// for comman users.
|
||||
void set_passes_internal(
|
||||
const std::vector<std::string> &passes_internal = {}) {
|
||||
passes_internal_ = passes_internal;
|
||||
}
|
||||
|
||||
const std::vector<std::string> &get_passes_internal() const {
|
||||
return passes_internal_;
|
||||
}
|
||||
|
||||
const std::vector<Place> &valid_places() const { return valid_places_; }
|
||||
|
||||
std::string model_file() const { return model_file_; }
|
||||
|
||||
std::string param_file() const { return param_file_; }
|
||||
|
||||
bool is_model_from_memory() const { return static_cast<bool>(model_buffer_); }
|
||||
|
||||
// note: `model_from_memory` has the same effect as `is_model_from_memory`,
|
||||
// but is_model_from_memory is recommended and `model_from_memory` will be
|
||||
// abandoned in v3.0.
|
||||
bool model_from_memory() const { return static_cast<bool>(model_buffer_); }
|
||||
public:
|
||||
void set_valid_places(const std::vector<Place>& x) { valid_places_ = x; }
|
||||
void set_model_file(const std::string& path) { model_file_ = path; }
|
||||
void set_param_file(const std::string& path) { param_file_ = path; }
|
||||
void set_model_buffer(const char* model_buffer,
|
||||
size_t model_buffer_size,
|
||||
const char* param_buffer,
|
||||
size_t param_buffer_size) {
|
||||
model_buffer_.reset(new CxxModelBuffer(
|
||||
model_buffer, model_buffer_size, param_buffer, param_buffer_size));
|
||||
}
|
||||
void set_model_buffer(std::shared_ptr<CxxModelBuffer> model_buffer) {
|
||||
model_buffer_ = model_buffer;
|
||||
}
|
||||
const CxxModelBuffer& get_model_buffer() const;
|
||||
// internal inference to choose passes for model optimizing,
|
||||
// it's designed for internal developer and not recommanded
|
||||
// for comman users.
|
||||
void set_passes_internal(
|
||||
const std::vector<std::string>& passes_internal = {}) {
|
||||
passes_internal_ = passes_internal;
|
||||
}
|
||||
const std::vector<std::string>& get_passes_internal() const {
|
||||
return passes_internal_;
|
||||
}
|
||||
const std::vector<Place>& valid_places() const { return valid_places_; }
|
||||
std::string model_file() const { return model_file_; }
|
||||
std::string param_file() const { return param_file_; }
|
||||
bool is_model_from_memory() const { return static_cast<bool>(model_buffer_); }
|
||||
// note: `model_from_memory` has the same effect as `is_model_from_memory`,
|
||||
// but is_model_from_memory is recommended and `model_from_memory` will be
|
||||
// abandoned in v3.0.
|
||||
bool model_from_memory() const { return static_cast<bool>(model_buffer_); }
|
||||
|
||||
#ifdef LITE_WITH_CUDA
|
||||
void set_multi_stream(bool multi_stream) { multi_stream_ = multi_stream; }
|
||||
bool multi_stream() const { return multi_stream_; }
|
||||
void set_multi_stream(bool multi_stream) { multi_stream_ = multi_stream; }
|
||||
bool multi_stream() const { return multi_stream_; }
|
||||
#endif
|
||||
|
||||
#ifdef LITE_WITH_MLU
|
||||
// set MLU core version, which is used when compiling MLU kernels
|
||||
void set_mlu_core_version(lite_api::MLUCoreVersion core_version);
|
||||
// set MLU core number, which is used when compiling MLU kernels
|
||||
void set_mlu_core_number(int core_number);
|
||||
// whether use MLU's first conv kernel. First conv is a special kernel
|
||||
// provided by MLU, its input is uint8, and also needs two 3-dimentional
|
||||
// vectors which save all inputs' mean and std values
|
||||
// set the 3-dimentional mean vector and 3-dimentional std vector used by
|
||||
// MLU's first conv
|
||||
void set_mlu_firstconv_param(const std::vector<float>& mean,
|
||||
const std::vector<float>& std);
|
||||
// set MLU input layout. User can specify layout of input data to be NHWC,
|
||||
// default is NCHW
|
||||
void set_mlu_input_layout(DataLayoutType layout);
|
||||
// set MLU core version, which is used when compiling MLU kernels
|
||||
void set_mlu_core_version(lite_api::MLUCoreVersion core_version);
|
||||
// set MLU core number, which is used when compiling MLU kernels
|
||||
void set_mlu_core_number(int core_number);
|
||||
// whether use MLU's first conv kernel. First conv is a special kernel
|
||||
// provided by MLU, its input is uint8, and also needs two 3-dimentional
|
||||
// vectors which save all inputs' mean and std values
|
||||
// set the 3-dimentional mean vector and 3-dimentional std vector used by
|
||||
// MLU's first conv
|
||||
void set_mlu_firstconv_param(const std::vector<float>& mean,
|
||||
const std::vector<float>& std);
|
||||
// set MLU input layout. User can specify layout of input data to be NHWC,
|
||||
// default is NCHW
|
||||
void set_mlu_input_layout(DataLayoutType layout);
|
||||
|
||||
lite_api::MLUCoreVersion mlu_core_version() const;
|
||||
int mlu_core_number() const;
|
||||
DataLayoutType mlu_input_layout() const;
|
||||
// std::pair<mean, std>
|
||||
std::pair<std::vector<float>, std::vector<float>> mlu_firstconv_param() const;
|
||||
lite_api::MLUCoreVersion mlu_core_version() const;
|
||||
int mlu_core_number() const;
|
||||
DataLayoutType mlu_input_layout() const;
|
||||
// std::pair<mean, std>
|
||||
std::pair<std::vector<float>, std::vector<float>> mlu_firstconv_param() const;
|
||||
#endif
|
||||
|
||||
// XPU only, set the size of the workspace memory from L3 cache for the
|
||||
// current thread.
|
||||
// **DEPRECATED**, use set_xpu_l3_cache_method() in the future
|
||||
void set_xpu_workspace_l3_size_per_thread(int l3_size = 0x4000000);
|
||||
// XPU only, set the size of the workspace memory from L3 cache for the
|
||||
// current thread.
|
||||
// **DEPRECATED**, use set_xpu_l3_cache_method() in the future
|
||||
void set_xpu_workspace_l3_size_per_thread(int l3_size = 0x4000000);
|
||||
void set_xpu_l3_cache_method(size_t l3_size, bool locked = false);
|
||||
|
||||
void set_xpu_l3_cache_method(size_t l3_size, bool locked = false);
|
||||
void set_xpu_gm_workspace_method(size_t gm_size);
|
||||
|
||||
void set_xpu_gm_workspace_method(size_t gm_size);
|
||||
void set_xpu_conv_autotune(bool autotune = true,
|
||||
const std::string& autotune_file = "");
|
||||
|
||||
void set_xpu_conv_autotune(bool autotune = true,
|
||||
const std::string &autotune_file = "");
|
||||
// XPU only, specify the target device ID for the current thread.
|
||||
// **DEPRECATED**, use xpu_set_device() at the very beginning of each worker
|
||||
// thread
|
||||
void set_xpu_dev_per_thread(int dev_no = 0);
|
||||
|
||||
// XPU only, specify the target device ID for the current thread.
|
||||
// **DEPRECATED**, use xpu_set_device() at the very beginning of each worker
|
||||
// thread
|
||||
void set_xpu_dev_per_thread(int dev_no = 0);
|
||||
// **DEPRECATED**, use set_xpu_multi_encoder_method() in the future
|
||||
void set_xpu_multi_encoder_precision(const std::string& precision = "int16");
|
||||
void set_xpu_multi_encoder_method(const std::string& precision = "int16",
|
||||
bool adaptive_seqlen = false);
|
||||
|
||||
// **DEPRECATED**, use set_xpu_multi_encoder_method() in the future
|
||||
void set_xpu_multi_encoder_precision(const std::string &precision = "int16");
|
||||
// set input tensor for warmup.
|
||||
// It is optional. If you set prefered_inputs, model wil run immediately when
|
||||
// predictor is created
|
||||
template <class T>
|
||||
void set_preferred_inputs_for_warmup(const int group_idx,
|
||||
const int tensor_idx,
|
||||
const shape_t& shape,
|
||||
const lod_t& lod = {},
|
||||
const T fill_value = 0,
|
||||
const void* data = nullptr);
|
||||
const std::map<int, std::vector<std::shared_ptr<void>>>&
|
||||
preferred_inputs_for_warmup() const {
|
||||
return preferred_inputs_for_warmup_;
|
||||
}
|
||||
|
||||
void set_xpu_multi_encoder_method(const std::string &precision = "int16",
|
||||
bool adaptive_seqlen = false);
|
||||
void set_quant_model(bool quant_model) { quant_model_ = quant_model; }
|
||||
bool quant_model() const { return quant_model_; }
|
||||
void set_quant_type(QuantType quant_type) { quant_type_ = quant_type; }
|
||||
QuantType quant_type() const { return quant_type_; }
|
||||
|
||||
// set input tensor for warmup.
|
||||
// It is optional. If you set prefered_inputs, model wil run immediately when
|
||||
// predictor is created
|
||||
template<class T>
|
||||
void set_preferred_inputs_for_warmup(const int group_idx,
|
||||
const int tensor_idx,
|
||||
const shape_t &shape,
|
||||
const lod_t &lod = {},
|
||||
const T fill_value = 0,
|
||||
const void *data = nullptr);
|
||||
|
||||
const std::map<int, std::vector<std::shared_ptr<void>>> &
|
||||
preferred_inputs_for_warmup() const {
|
||||
return preferred_inputs_for_warmup_;
|
||||
}
|
||||
|
||||
void set_quant_model(bool quant_model) { quant_model_ = quant_model; }
|
||||
|
||||
bool quant_model() const { return quant_model_; }
|
||||
|
||||
void set_quant_type(QuantType quant_type) { quant_type_ = quant_type; }
|
||||
|
||||
QuantType quant_type() const { return quant_type_; }
|
||||
|
||||
void set_sparse_model(bool sparse_model) { sparse_model_ = sparse_model; }
|
||||
|
||||
bool sparse_model() const { return sparse_model_; }
|
||||
|
||||
void set_sparse_threshold(float sparse_threshold) {
|
||||
sparse_threshold_ = sparse_threshold;
|
||||
}
|
||||
|
||||
float sparse_threshold() const { return sparse_threshold_; }
|
||||
};
|
||||
void set_sparse_model(bool sparse_model) { sparse_model_ = sparse_model; }
|
||||
bool sparse_model() const { return sparse_model_; }
|
||||
void set_sparse_threshold(float sparse_threshold) {
|
||||
sparse_threshold_ = sparse_threshold;
|
||||
}
|
||||
float sparse_threshold() const { return sparse_threshold_; }
|
||||
};
|
||||
|
||||
/// MobileConfig is the config for the light weight predictor, it will skip
|
||||
/// IR optimization or other unnecessary stages.
|
||||
class LITE_API MobileConfig : public ConfigBase {
|
||||
// whether to load data from memory. Model data will be loaded from memory
|
||||
// buffer if model_from_memory_ is true.
|
||||
bool model_from_memory_{false};
|
||||
class LITE_API MobileConfig : public ConfigBase {
|
||||
// whether to load data from memory. Model data will be loaded from memory
|
||||
// buffer if model_from_memory_ is true.
|
||||
bool model_from_memory_{false};
|
||||
|
||||
// model data readed from file or memory buffer in combined format.
|
||||
std::string lite_model_file_;
|
||||
// model data readed from file or memory buffer in combined format.
|
||||
std::string lite_model_file_;
|
||||
|
||||
// NOTE: This is a deprecated variable and will be removed in latter release.
|
||||
std::string model_buffer_;
|
||||
std::string param_buffer_;
|
||||
// NOTE: This is a deprecated variable and will be removed in latter release.
|
||||
std::string model_buffer_;
|
||||
std::string param_buffer_;
|
||||
|
||||
public:
|
||||
// set model data in combined format, `set_model_from_file` refers to loading
|
||||
// model from file, set_model_from_buffer refers to loading model from memory
|
||||
// buffer
|
||||
void set_model_from_file(const std::string &x);
|
||||
public:
|
||||
// set model data in combined format, `set_model_from_file` refers to loading
|
||||
// model from file, set_model_from_buffer refers to loading model from memory
|
||||
// buffer
|
||||
void set_model_from_file(const std::string& x);
|
||||
void set_model_from_buffer(const std::string& x);
|
||||
// return model data in lite_model_file_, which is in combined format.
|
||||
const std::string& lite_model_file() const { return lite_model_file_; }
|
||||
|
||||
void set_model_from_buffer(const std::string &x);
|
||||
// return model_from_memory_, which indicates whether to load model from
|
||||
// memory buffer.
|
||||
bool is_model_from_memory() const { return model_from_memory_; }
|
||||
// note: `model_from_memory` has the same effect as `is_model_from_memory`,
|
||||
// but is_model_from_memory is recommended and `model_from_memory` will be
|
||||
// abandoned in v3.0.
|
||||
bool model_from_memory() const { return model_from_memory_; }
|
||||
|
||||
// return model data in lite_model_file_, which is in combined format.
|
||||
const std::string &lite_model_file() const { return lite_model_file_; }
|
||||
// NOTE: This is a deprecated API and will be removed in latter release.
|
||||
void set_model_buffer(const char* model_buffer,
|
||||
size_t model_buffer_size,
|
||||
const char* param_buffer,
|
||||
size_t param_buffer_size);
|
||||
|
||||
// return model_from_memory_, which indicates whether to load model from
|
||||
// memory buffer.
|
||||
bool is_model_from_memory() const { return model_from_memory_; }
|
||||
// NOTE: This is a deprecated API and will be removed in latter release.
|
||||
const std::string& model_buffer() const { return model_buffer_; }
|
||||
|
||||
// note: `model_from_memory` has the same effect as `is_model_from_memory`,
|
||||
// but is_model_from_memory is recommended and `model_from_memory` will be
|
||||
// abandoned in v3.0.
|
||||
bool model_from_memory() const { return model_from_memory_; }
|
||||
// NOTE: This is a deprecated API and will be removed in latter release.
|
||||
const std::string& param_buffer() const { return param_buffer_; }
|
||||
|
||||
// NOTE: This is a deprecated API and will be removed in latter release.
|
||||
void set_model_buffer(const char *model_buffer,
|
||||
size_t model_buffer_size,
|
||||
const char *param_buffer,
|
||||
size_t param_buffer_size);
|
||||
// This is the method for allocating workspace_size according to L3Cache size
|
||||
void SetArmL3CacheSize(
|
||||
L3CacheSetMethod method = L3CacheSetMethod::kDeviceL3Cache,
|
||||
int absolute_val = -1);
|
||||
};
|
||||
|
||||
// NOTE: This is a deprecated API and will be removed in latter release.
|
||||
const std::string &model_buffer() const { return model_buffer_; }
|
||||
template <typename ConfigT>
|
||||
LITE_API std::shared_ptr<PaddlePredictor> CreatePaddlePredictor(const ConfigT&);
|
||||
|
||||
// NOTE: This is a deprecated API and will be removed in latter release.
|
||||
const std::string ¶m_buffer() const { return param_buffer_; }
|
||||
|
||||
// This is the method for allocating workspace_size according to L3Cache size
|
||||
void SetArmL3CacheSize(
|
||||
L3CacheSetMethod method = L3CacheSetMethod::kDeviceL3Cache,
|
||||
int absolute_val = -1);
|
||||
};
|
||||
|
||||
template<typename ConfigT>
|
||||
LITE_API std::shared_ptr<PaddlePredictor> CreatePaddlePredictor(const ConfigT &);
|
||||
|
||||
} // namespace lite_api
|
||||
} // namespace lite_api
|
||||
} // namespace paddle
|
||||
|
||||
#endif // NOLINT
|
||||
|
||||
@@ -21,254 +21,254 @@
|
||||
#include "lite/api/paddle_place.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace lite {
|
||||
namespace utils {
|
||||
namespace cv {
|
||||
typedef paddle::lite_api::Tensor Tensor;
|
||||
typedef paddle::lite_api::DataLayoutType LayoutType;
|
||||
namespace lite {
|
||||
namespace utils {
|
||||
namespace cv {
|
||||
typedef paddle::lite_api::Tensor Tensor;
|
||||
typedef paddle::lite_api::DataLayoutType LayoutType;
|
||||
// color enum
|
||||
enum ImageFormat {
|
||||
RGBA = 0,
|
||||
BGRA,
|
||||
RGB,
|
||||
BGR,
|
||||
GRAY,
|
||||
NV21 = 11,
|
||||
NV12,
|
||||
YUV420SP,
|
||||
YUV420P,
|
||||
YUV422,
|
||||
YUV444
|
||||
};
|
||||
enum ImageFormat {
|
||||
RGBA = 0,
|
||||
BGRA,
|
||||
RGB,
|
||||
BGR,
|
||||
GRAY,
|
||||
NV21 = 11,
|
||||
NV12,
|
||||
YUV420SP,
|
||||
YUV420P,
|
||||
YUV422,
|
||||
YUV444
|
||||
};
|
||||
// flip enum
|
||||
enum FlipParam {
|
||||
XY = -1, // flip along the XY axis
|
||||
X = 0, // flip along the X axis
|
||||
Y // flip along the Y axis
|
||||
};
|
||||
enum FlipParam {
|
||||
XY = -1, // flip along the XY axis
|
||||
X = 0, // flip along the X axis
|
||||
Y // flip along the Y axis
|
||||
};
|
||||
// transform param
|
||||
typedef struct {
|
||||
int ih; // input height
|
||||
int iw; // input width
|
||||
int oh; // outpu theight
|
||||
int ow; // output width
|
||||
FlipParam flip_param; // flip, support x, y, xy
|
||||
float rotate_param; // rotate, support 90, 180, 270
|
||||
} TransParam;
|
||||
typedef struct {
|
||||
int ih; // input height
|
||||
int iw; // input width
|
||||
int oh; // outpu theight
|
||||
int ow; // output width
|
||||
FlipParam flip_param; // flip, support x, y, xy
|
||||
float rotate_param; // rotate, support 90, 180, 270
|
||||
} TransParam;
|
||||
|
||||
class ImagePreprocess {
|
||||
public:
|
||||
/*
|
||||
* init
|
||||
* param srcFormat: input image color
|
||||
* param dstFormat: output image color
|
||||
* param param: input image parameter, egs: input size
|
||||
*/
|
||||
ImagePreprocess(ImageFormat srcFormat,
|
||||
ImageFormat dstFormat,
|
||||
TransParam param);
|
||||
class ImagePreprocess {
|
||||
public:
|
||||
/*
|
||||
* init
|
||||
* param srcFormat: input image color
|
||||
* param dstFormat: output image color
|
||||
* param param: input image parameter, egs: input size
|
||||
*/
|
||||
ImagePreprocess(ImageFormat srcFormat,
|
||||
ImageFormat dstFormat,
|
||||
TransParam param);
|
||||
|
||||
/*
|
||||
* image color convert
|
||||
* support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
|
||||
* BGR(RGB)and BGRA(RGBA) transform,
|
||||
* BGR(RGB)and RGB(BGR) transform,
|
||||
* BGR(RGB)and RGBA(BGRA) transform,
|
||||
* BGR(RGB) and GRAY transform,
|
||||
* BGRA(RGBA) and GRAY transform,
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_convert(const uint8_t *src, uint8_t *dst);
|
||||
/*
|
||||
* image color convert
|
||||
* support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
|
||||
* BGR(RGB)and BGRA(RGBA) transform,
|
||||
* BGR(RGB)and RGB(BGR) transform,
|
||||
* BGR(RGB)and RGBA(BGRA) transform,
|
||||
* BGR(RGB) and GRAY transform,
|
||||
* BGRA(RGBA) and GRAY transform,
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_convert(const uint8_t* src, uint8_t* dst);
|
||||
|
||||
/*
|
||||
* image color convert
|
||||
* support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
|
||||
* BGR(RGB)and BGRA(RGBA) transform,
|
||||
* BGR(RGB)and RGB(BGR) transform,
|
||||
* BGR(RGB)and RGBA(BGRA) transform,
|
||||
* BGR(RGB)and GRAY transform,
|
||||
* BGRA(RGBA) and GRAY transform,
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image image format support: GRAY, NV12(NV21),
|
||||
* BGR(RGB) and BGRA(RGBA)
|
||||
* param dstFormat: output image image format, support GRAY, BGR(RGB) and
|
||||
* BGRA(RGBA)
|
||||
*/
|
||||
void image_convert(const uint8_t *src,
|
||||
uint8_t *dst,
|
||||
ImageFormat srcFormat,
|
||||
ImageFormat dstFormat);
|
||||
/*
|
||||
* image color convert
|
||||
* support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
|
||||
* BGR(RGB)and BGRA(RGBA) transform,
|
||||
* BGR(RGB)and RGB(BGR) transform,
|
||||
* BGR(RGB)and RGBA(BGRA) transform,
|
||||
* BGR(RGB)and GRAY transform,
|
||||
* BGRA(RGBA) and GRAY transform,
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image image format support: GRAY, NV12(NV21),
|
||||
* BGR(RGB) and BGRA(RGBA)
|
||||
* param dstFormat: output image image format, support GRAY, BGR(RGB) and
|
||||
* BGRA(RGBA)
|
||||
*/
|
||||
void image_convert(const uint8_t* src,
|
||||
uint8_t* dst,
|
||||
ImageFormat srcFormat,
|
||||
ImageFormat dstFormat);
|
||||
|
||||
/*
|
||||
* image color convert
|
||||
* support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
|
||||
* BGR(RGB)and BGRA(RGBA) transform,
|
||||
* BGR(RGB)and RGB(BGR) transform,
|
||||
* BGR(RGB)and RGBA(BGRA) transform,
|
||||
* BGR(RGB)and GRAY transform,
|
||||
* BGRA(RGBA) and GRAY transform,
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image image format support: GRAY, NV12(NV21),
|
||||
* BGR(RGB) and BGRA(RGBA)
|
||||
* param dstFormat: output image image format, support GRAY, BGR(RGB) and
|
||||
* BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
*/
|
||||
void image_convert(const uint8_t *src,
|
||||
uint8_t *dst,
|
||||
ImageFormat srcFormat,
|
||||
ImageFormat dstFormat,
|
||||
int srcw,
|
||||
int srch);
|
||||
/*
|
||||
* image color convert
|
||||
* support NV12/NV21_to_BGR(RGB), NV12/NV21_to_BGRA(RGBA),
|
||||
* BGR(RGB)and BGRA(RGBA) transform,
|
||||
* BGR(RGB)and RGB(BGR) transform,
|
||||
* BGR(RGB)and RGBA(BGRA) transform,
|
||||
* BGR(RGB)and GRAY transform,
|
||||
* BGRA(RGBA) and GRAY transform,
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image image format support: GRAY, NV12(NV21),
|
||||
* BGR(RGB) and BGRA(RGBA)
|
||||
* param dstFormat: output image image format, support GRAY, BGR(RGB) and
|
||||
* BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
*/
|
||||
void image_convert(const uint8_t* src,
|
||||
uint8_t* dst,
|
||||
ImageFormat srcFormat,
|
||||
ImageFormat dstFormat,
|
||||
int srcw,
|
||||
int srch);
|
||||
|
||||
/*
|
||||
* image resize, use bilinear method
|
||||
* support image format: 1-channel image (egs: GRAY, 2-channel image (egs:
|
||||
* NV12, NV21), 3-channel(egs: BGR), 4-channel(egs: BGRA)
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_resize(const uint8_t *src, uint8_t *dst);
|
||||
/*
|
||||
* image resize, use bilinear method
|
||||
* support image format: 1-channel image (egs: GRAY, 2-channel image (egs:
|
||||
* NV12, NV21), 3-channel(egs: BGR), 4-channel(egs: BGRA)
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_resize(const uint8_t* src, uint8_t* dst);
|
||||
|
||||
/*
|
||||
image resize, use bilinear method
|
||||
* support image format: 1-channel image (egs: GRAY, 2-channel image (egs:
|
||||
NV12, NV21), 3-channel image(egs: BGR), 4-channel image(egs: BGRA)
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param dstw: output image width
|
||||
* param dsth: output image height
|
||||
*/
|
||||
void image_resize(const uint8_t *src,
|
||||
uint8_t *dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
int dstw,
|
||||
int dsth);
|
||||
/*
|
||||
image resize, use bilinear method
|
||||
* support image format: 1-channel image (egs: GRAY, 2-channel image (egs:
|
||||
NV12, NV21), 3-channel image(egs: BGR), 4-channel image(egs: BGRA)
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param dstw: output image width
|
||||
* param dsth: output image height
|
||||
*/
|
||||
void image_resize(const uint8_t* src,
|
||||
uint8_t* dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
int dstw,
|
||||
int dsth);
|
||||
|
||||
/*
|
||||
* image Rotate
|
||||
* support 90, 180 and 270 Rotate process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_rotate(const uint8_t *src, uint8_t *dst);
|
||||
/*
|
||||
* image Rotate
|
||||
* support 90, 180 and 270 Rotate process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_rotate(const uint8_t* src, uint8_t* dst);
|
||||
|
||||
/*
|
||||
* image Rotate
|
||||
* support 90, 180 and 270 Rotate process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image format, support GRAY, BGR(RGB) and BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param degree: Rotate degree, support 90, 180 and 270
|
||||
*/
|
||||
void image_rotate(const uint8_t *src,
|
||||
uint8_t *dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
float degree);
|
||||
/*
|
||||
* image Rotate
|
||||
* support 90, 180 and 270 Rotate process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image format, support GRAY, BGR(RGB) and BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param degree: Rotate degree, support 90, 180 and 270
|
||||
*/
|
||||
void image_rotate(const uint8_t* src,
|
||||
uint8_t* dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
float degree);
|
||||
|
||||
/*
|
||||
* image Flip
|
||||
* support X, Y and XY flip process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_flip(const uint8_t *src, uint8_t *dst);
|
||||
/*
|
||||
* image Flip
|
||||
* support X, Y and XY flip process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_flip(const uint8_t* src, uint8_t* dst);
|
||||
|
||||
/*
|
||||
* image Flip
|
||||
* support X, Y and XY flip process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image format, support GRAY, BGR(RGB) and BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param flip_param: flip parameter, support X, Y and XY
|
||||
*/
|
||||
void image_flip(const uint8_t *src,
|
||||
uint8_t *dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
FlipParam flip_param);
|
||||
/*
|
||||
* image Flip
|
||||
* support X, Y and XY flip process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
* param srcFormat: input image format, support GRAY, BGR(RGB) and BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param flip_param: flip parameter, support X, Y and XY
|
||||
*/
|
||||
void image_flip(const uint8_t* src,
|
||||
uint8_t* dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
FlipParam flip_param);
|
||||
|
||||
/*
|
||||
* change image data to tensor data
|
||||
* support image format is GRAY, BGR(RGB) and BGRA(RGBA), Data layout is NHWC
|
||||
* and
|
||||
* NCHW
|
||||
* param src: input image data
|
||||
* param dstTensor: output tensor data
|
||||
* param layout: output tensor layout，support NHWC and NCHW
|
||||
* param means: means of image
|
||||
* param scales: scales of image
|
||||
*/
|
||||
void image_to_tensor(const uint8_t *src,
|
||||
Tensor *dstTensor,
|
||||
LayoutType layout,
|
||||
float *means,
|
||||
float *scales);
|
||||
/*
|
||||
* change image data to tensor data
|
||||
* support image format is GRAY, BGR(RGB) and BGRA(RGBA), Data layout is NHWC
|
||||
* and
|
||||
* NCHW
|
||||
* param src: input image data
|
||||
* param dstTensor: output tensor data
|
||||
* param layout: output tensor layout，support NHWC and NCHW
|
||||
* param means: means of image
|
||||
* param scales: scales of image
|
||||
*/
|
||||
void image_to_tensor(const uint8_t* src,
|
||||
Tensor* dstTensor,
|
||||
LayoutType layout,
|
||||
float* means,
|
||||
float* scales);
|
||||
|
||||
/*
|
||||
* change image data to tensor data
|
||||
* support image format is GRAY, BGR(RGB) and BGRA(RGBA), Data layout is NHWC
|
||||
* and
|
||||
* NCHW
|
||||
* param src: input image data
|
||||
* param dstTensor: output tensor data
|
||||
* param srcFormat: input image format, support BGR(RGB) and BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param layout: output tensor layout，support NHWC and NCHW
|
||||
* param means: means of image
|
||||
* param scales: scales of image
|
||||
*/
|
||||
void image_to_tensor(const uint8_t *src,
|
||||
Tensor *dstTensor,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
LayoutType layout,
|
||||
float *means,
|
||||
float *scales);
|
||||
/*
|
||||
* change image data to tensor data
|
||||
* support image format is GRAY, BGR(RGB) and BGRA(RGBA), Data layout is NHWC
|
||||
* and
|
||||
* NCHW
|
||||
* param src: input image data
|
||||
* param dstTensor: output tensor data
|
||||
* param srcFormat: input image format, support BGR(RGB) and BGRA(RGBA)
|
||||
* param srcw: input image width
|
||||
* param srch: input image height
|
||||
* param layout: output tensor layout，support NHWC and NCHW
|
||||
* param means: means of image
|
||||
* param scales: scales of image
|
||||
*/
|
||||
void image_to_tensor(const uint8_t* src,
|
||||
Tensor* dstTensor,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
LayoutType layout,
|
||||
float* means,
|
||||
float* scales);
|
||||
|
||||
/*
|
||||
* image crop process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_crop(const uint8_t *src,
|
||||
uint8_t *dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
int left_x,
|
||||
int left_y,
|
||||
int dstw,
|
||||
int dsth);
|
||||
/*
|
||||
* image crop process
|
||||
* color format support 1-channel image, 3-channel image and 4-channel image
|
||||
* param src: input image data
|
||||
* param dst: output image data
|
||||
*/
|
||||
void image_crop(const uint8_t* src,
|
||||
uint8_t* dst,
|
||||
ImageFormat srcFormat,
|
||||
int srcw,
|
||||
int srch,
|
||||
int left_x,
|
||||
int left_y,
|
||||
int dstw,
|
||||
int dsth);
|
||||
|
||||
private:
|
||||
ImageFormat srcFormat_;
|
||||
ImageFormat dstFormat_;
|
||||
TransParam transParam_;
|
||||
};
|
||||
} // namespace cv
|
||||
} // namespace utils
|
||||
} // namespace lite
|
||||
private:
|
||||
ImageFormat srcFormat_;
|
||||
ImageFormat dstFormat_;
|
||||
TransParam transParam_;
|
||||
};
|
||||
} // namespace cv
|
||||
} // namespace utils
|
||||
} // namespace lite
|
||||
} // namespace paddle
|
||||
|
||||
@@ -13,7 +13,6 @@
|
||||
// limitations under the License.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <set>
|
||||
#include <string>
|
||||
|
||||
@@ -41,142 +40,140 @@
|
||||
#endif
|
||||
|
||||
namespace paddle {
|
||||
namespace lite_api {
|
||||
namespace lite_api {
|
||||
|
||||
enum class TargetType : int {
|
||||
kUnk = 0,
|
||||
kHost = 1,
|
||||
kX86 = 2,
|
||||
kCUDA = 3,
|
||||
kARM = 4,
|
||||
kOpenCL = 5,
|
||||
kAny = 6, // any target
|
||||
kFPGA = 7,
|
||||
kNPU = 8,
|
||||
kXPU = 9,
|
||||
kBM = 10,
|
||||
kMLU = 11,
|
||||
kRKNPU = 12,
|
||||
kAPU = 13,
|
||||
kHuaweiAscendNPU = 14,
|
||||
kImaginationNNA = 15,
|
||||
kIntelFPGA = 16,
|
||||
kMetal = 17,
|
||||
kNNAdapter = 18,
|
||||
NUM = 19, // number of fields.
|
||||
};
|
||||
enum class PrecisionType : int {
|
||||
kUnk = 0,
|
||||
kFloat = 1,
|
||||
kInt8 = 2,
|
||||
kInt32 = 3,
|
||||
kAny = 4, // any precision
|
||||
kFP16 = 5,
|
||||
kBool = 6,
|
||||
kInt64 = 7,
|
||||
kInt16 = 8,
|
||||
kUInt8 = 9,
|
||||
kFP64 = 10,
|
||||
NUM = 11, // number of fields.
|
||||
};
|
||||
enum class DataLayoutType : int {
|
||||
kUnk = 0,
|
||||
kNCHW = 1,
|
||||
kNHWC = 3,
|
||||
kImageDefault = 4, // for opencl image2d
|
||||
kImageFolder = 5, // for opencl image2d
|
||||
kImageNW = 6, // for opencl image2d
|
||||
kAny = 2, // any data layout
|
||||
kMetalTexture2DArray = 7,
|
||||
kMetalTexture2D = 8,
|
||||
NUM = 9, // number of fields.
|
||||
};
|
||||
enum class TargetType : int {
|
||||
kUnk = 0,
|
||||
kHost = 1,
|
||||
kX86 = 2,
|
||||
kCUDA = 3,
|
||||
kARM = 4,
|
||||
kOpenCL = 5,
|
||||
kAny = 6, // any target
|
||||
kFPGA = 7,
|
||||
kNPU = 8,
|
||||
kXPU = 9,
|
||||
kBM = 10,
|
||||
kMLU = 11,
|
||||
kRKNPU = 12,
|
||||
kAPU = 13,
|
||||
kHuaweiAscendNPU = 14,
|
||||
kImaginationNNA = 15,
|
||||
kIntelFPGA = 16,
|
||||
kMetal = 17,
|
||||
kNNAdapter = 18,
|
||||
NUM = 19, // number of fields.
|
||||
};
|
||||
enum class PrecisionType : int {
|
||||
kUnk = 0,
|
||||
kFloat = 1,
|
||||
kInt8 = 2,
|
||||
kInt32 = 3,
|
||||
kAny = 4, // any precision
|
||||
kFP16 = 5,
|
||||
kBool = 6,
|
||||
kInt64 = 7,
|
||||
kInt16 = 8,
|
||||
kUInt8 = 9,
|
||||
kFP64 = 10,
|
||||
NUM = 11, // number of fields.
|
||||
};
|
||||
enum class DataLayoutType : int {
|
||||
kUnk = 0,
|
||||
kNCHW = 1,
|
||||
kNHWC = 3,
|
||||
kImageDefault = 4, // for opencl image2d
|
||||
kImageFolder = 5, // for opencl image2d
|
||||
kImageNW = 6, // for opencl image2d
|
||||
kAny = 2, // any data layout
|
||||
kMetalTexture2DArray = 7,
|
||||
kMetalTexture2D = 8,
|
||||
NUM = 9, // number of fields.
|
||||
};
|
||||
|
||||
typedef enum {
|
||||
LITE_POWER_HIGH = 0,
|
||||
LITE_POWER_LOW = 1,
|
||||
LITE_POWER_FULL = 2,
|
||||
LITE_POWER_NO_BIND = 3,
|
||||
LITE_POWER_RAND_HIGH = 4,
|
||||
LITE_POWER_RAND_LOW = 5
|
||||
} PowerMode;
|
||||
typedef enum {
|
||||
LITE_POWER_HIGH = 0,
|
||||
LITE_POWER_LOW = 1,
|
||||
LITE_POWER_FULL = 2,
|
||||
LITE_POWER_NO_BIND = 3,
|
||||
LITE_POWER_RAND_HIGH = 4,
|
||||
LITE_POWER_RAND_LOW = 5
|
||||
} PowerMode;
|
||||
|
||||
typedef enum {
|
||||
CL_TUNE_NONE = 0,
|
||||
CL_TUNE_RAPID = 1,
|
||||
CL_TUNE_NORMAL = 2,
|
||||
CL_TUNE_EXHAUSTIVE = 3
|
||||
} CLTuneMode;
|
||||
typedef enum {
|
||||
CL_TUNE_NONE = 0,
|
||||
CL_TUNE_RAPID = 1,
|
||||
CL_TUNE_NORMAL = 2,
|
||||
CL_TUNE_EXHAUSTIVE = 3
|
||||
} CLTuneMode;
|
||||
|
||||
typedef enum {
|
||||
CL_PRECISION_AUTO = 0,
|
||||
CL_PRECISION_FP32 = 1,
|
||||
CL_PRECISION_FP16 = 2
|
||||
} CLPrecisionType;
|
||||
typedef enum {
|
||||
CL_PRECISION_AUTO = 0,
|
||||
CL_PRECISION_FP32 = 1,
|
||||
CL_PRECISION_FP16 = 2
|
||||
} CLPrecisionType;
|
||||
|
||||
typedef enum {
|
||||
MLU_220 = 0, MLU_270 = 1
|
||||
} MLUCoreVersion;
|
||||
typedef enum { MLU_220 = 0, MLU_270 = 1 } MLUCoreVersion;
|
||||
|
||||
enum class ActivationType : int {
|
||||
kIndentity = 0,
|
||||
kRelu = 1,
|
||||
kRelu6 = 2,
|
||||
kPRelu = 3,
|
||||
kLeakyRelu = 4,
|
||||
kSigmoid = 5,
|
||||
kTanh = 6,
|
||||
kSwish = 7,
|
||||
kExp = 8,
|
||||
kAbs = 9,
|
||||
kHardSwish = 10,
|
||||
kReciprocal = 11,
|
||||
kThresholdedRelu = 12,
|
||||
kElu = 13,
|
||||
kHardSigmoid = 14,
|
||||
kLog = 15,
|
||||
kSigmoid_v2 = 16,
|
||||
kTanh_v2 = 17,
|
||||
kGelu = 18,
|
||||
kErf = 19,
|
||||
kSign = 20,
|
||||
kSoftPlus = 21,
|
||||
kMish = 22,
|
||||
NUM = 23,
|
||||
};
|
||||
enum class ActivationType : int {
|
||||
kIndentity = 0,
|
||||
kRelu = 1,
|
||||
kRelu6 = 2,
|
||||
kPRelu = 3,
|
||||
kLeakyRelu = 4,
|
||||
kSigmoid = 5,
|
||||
kTanh = 6,
|
||||
kSwish = 7,
|
||||
kExp = 8,
|
||||
kAbs = 9,
|
||||
kHardSwish = 10,
|
||||
kReciprocal = 11,
|
||||
kThresholdedRelu = 12,
|
||||
kElu = 13,
|
||||
kHardSigmoid = 14,
|
||||
kLog = 15,
|
||||
kSigmoid_v2 = 16,
|
||||
kTanh_v2 = 17,
|
||||
kGelu = 18,
|
||||
kErf = 19,
|
||||
kSign = 20,
|
||||
kSoftPlus = 21,
|
||||
kMish = 22,
|
||||
NUM = 23,
|
||||
};
|
||||
|
||||
static size_t PrecisionTypeLength(PrecisionType type) {
|
||||
switch (type) {
|
||||
case PrecisionType::kFloat:
|
||||
return 4;
|
||||
case PrecisionType::kFP64:
|
||||
return 8;
|
||||
case PrecisionType::kUInt8:
|
||||
return 1;
|
||||
case PrecisionType::kInt8:
|
||||
return 1;
|
||||
case PrecisionType::kInt32:
|
||||
return 4;
|
||||
case PrecisionType::kInt64:
|
||||
return 8;
|
||||
case PrecisionType::kFP16:
|
||||
return 2;
|
||||
case PrecisionType::kInt16:
|
||||
return 2;
|
||||
default:
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
static size_t PrecisionTypeLength(PrecisionType type) {
|
||||
switch (type) {
|
||||
case PrecisionType::kFloat:
|
||||
return 4;
|
||||
case PrecisionType::kFP64:
|
||||
return 8;
|
||||
case PrecisionType::kUInt8:
|
||||
return 1;
|
||||
case PrecisionType::kInt8:
|
||||
return 1;
|
||||
case PrecisionType::kInt32:
|
||||
return 4;
|
||||
case PrecisionType::kInt64:
|
||||
return 8;
|
||||
case PrecisionType::kFP16:
|
||||
return 2;
|
||||
case PrecisionType::kInt16:
|
||||
return 2;
|
||||
default:
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
enum class QuantType : int {
|
||||
QUANT_INT8,
|
||||
QUANT_INT16,
|
||||
};
|
||||
enum class QuantType : int {
|
||||
QUANT_INT8,
|
||||
QUANT_INT16,
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
struct PrecisionTypeTrait {
|
||||
constexpr static PrecisionType Type() { return PrecisionType::kUnk; }
|
||||
};
|
||||
template <typename T>
|
||||
struct PrecisionTypeTrait {
|
||||
constexpr static PrecisionType Type() { return PrecisionType::kUnk; }
|
||||
};
|
||||
|
||||
#define _ForEachPrecisionTypeHelper(callback, cpp_type, precision_type) \
|
||||
callback(cpp_type, ::paddle::lite_api::PrecisionType::precision_type);
|
||||
@@ -197,11 +194,11 @@ namespace paddle {
|
||||
constexpr static PrecisionType Type() { return precision_type; } \
|
||||
}
|
||||
|
||||
_ForEachPrecisionType(DefinePrecisionTypeTrait);
|
||||
_ForEachPrecisionType(DefinePrecisionTypeTrait);
|
||||
|
||||
#ifdef ENABLE_ARM_FP16
|
||||
typedef __fp16 float16_t;
|
||||
_ForEachPrecisionTypeHelper(DefinePrecisionTypeTrait, float16_t, kFP16);
|
||||
typedef __fp16 float16_t;
|
||||
_ForEachPrecisionTypeHelper(DefinePrecisionTypeTrait, float16_t, kFP16);
|
||||
#endif
|
||||
|
||||
#undef _ForEachPrecisionTypeHelper
|
||||
@@ -212,71 +209,70 @@ namespace paddle {
|
||||
#define PRECISION(item__) paddle::lite_api::PrecisionType::item__
|
||||
#define DATALAYOUT(item__) paddle::lite_api::DataLayoutType::item__
|
||||
|
||||
const std::string &ActivationTypeToStr(ActivationType act);
|
||||
const std::string& ActivationTypeToStr(ActivationType act);
|
||||
|
||||
const std::string &TargetToStr(TargetType target);
|
||||
const std::string& TargetToStr(TargetType target);
|
||||
|
||||
const std::string &PrecisionToStr(PrecisionType precision);
|
||||
const std::string& PrecisionToStr(PrecisionType precision);
|
||||
|
||||
const std::string &DataLayoutToStr(DataLayoutType layout);
|
||||
const std::string& DataLayoutToStr(DataLayoutType layout);
|
||||
|
||||
const std::string &TargetRepr(TargetType target);
|
||||
const std::string& TargetRepr(TargetType target);
|
||||
|
||||
const std::string &PrecisionRepr(PrecisionType precision);
|
||||
const std::string& PrecisionRepr(PrecisionType precision);
|
||||
|
||||
const std::string &DataLayoutRepr(DataLayoutType layout);
|
||||
const std::string& DataLayoutRepr(DataLayoutType layout);
|
||||
|
||||
const std::string &CLTuneModeToStr(CLTuneMode mode);
|
||||
const std::string& CLTuneModeToStr(CLTuneMode mode);
|
||||
|
||||
const std::string &CLPrecisionTypeToStr(CLPrecisionType type);
|
||||
const std::string& CLPrecisionTypeToStr(CLPrecisionType type);
|
||||
|
||||
// Get a set of all the elements represented by the target.
|
||||
std::set<TargetType> ExpandValidTargets(TargetType target = TARGET(kAny));
|
||||
std::set<TargetType> ExpandValidTargets(TargetType target = TARGET(kAny));
|
||||
|
||||
// Get a set of all the elements represented by the precision.
|
||||
std::set<PrecisionType> ExpandValidPrecisions(
|
||||
PrecisionType precision = PRECISION(kAny));
|
||||
std::set<PrecisionType> ExpandValidPrecisions(
|
||||
PrecisionType precision = PRECISION(kAny));
|
||||
|
||||
// Get a set of all the elements represented by the layout.
|
||||
std::set<DataLayoutType> ExpandValidLayouts(
|
||||
DataLayoutType layout = DATALAYOUT(kAny));
|
||||
std::set<DataLayoutType> ExpandValidLayouts(
|
||||
DataLayoutType layout = DATALAYOUT(kAny));
|
||||
|
||||
/*
|
||||
* Place specifies the execution context of a Kernel or input/output for a
|
||||
* kernel. It is used to make the analysis of the MIR more clear and accurate.
|
||||
*/
|
||||
struct LITE_API Place {
|
||||
TargetType target{TARGET(kUnk)};
|
||||
PrecisionType precision{PRECISION(kUnk)};
|
||||
DataLayoutType layout{DATALAYOUT(kUnk)};
|
||||
int16_t device{0}; // device ID
|
||||
struct LITE_API Place {
|
||||
TargetType target{TARGET(kUnk)};
|
||||
PrecisionType precision{PRECISION(kUnk)};
|
||||
DataLayoutType layout{DATALAYOUT(kUnk)};
|
||||
int16_t device{0}; // device ID
|
||||
|
||||
Place() = default;
|
||||
Place() = default;
|
||||
Place(TargetType target,
|
||||
PrecisionType precision = PRECISION(kFloat),
|
||||
DataLayoutType layout = DATALAYOUT(kNCHW),
|
||||
int16_t device = 0)
|
||||
: target(target), precision(precision), layout(layout), device(device) {}
|
||||
|
||||
Place(TargetType target,
|
||||
PrecisionType precision = PRECISION(kFloat),
|
||||
DataLayoutType layout = DATALAYOUT(kNCHW),
|
||||
int16_t device = 0)
|
||||
: target(target), precision(precision), layout(layout), device(device) {}
|
||||
bool is_valid() const {
|
||||
return target != TARGET(kUnk) && precision != PRECISION(kUnk) &&
|
||||
layout != DATALAYOUT(kUnk);
|
||||
}
|
||||
|
||||
bool is_valid() const {
|
||||
return target != TARGET(kUnk) && precision != PRECISION(kUnk) &&
|
||||
layout != DATALAYOUT(kUnk);
|
||||
}
|
||||
size_t hash() const;
|
||||
|
||||
size_t hash() const;
|
||||
bool operator==(const Place& other) const {
|
||||
return target == other.target && precision == other.precision &&
|
||||
layout == other.layout && device == other.device;
|
||||
}
|
||||
|
||||
bool operator==(const Place &other) const {
|
||||
return target == other.target && precision == other.precision &&
|
||||
layout == other.layout && device == other.device;
|
||||
}
|
||||
bool operator!=(const Place& other) const { return !(*this == other); }
|
||||
|
||||
bool operator!=(const Place &other) const { return !(*this == other); }
|
||||
friend bool operator<(const Place& a, const Place& b);
|
||||
|
||||
friend bool operator<(const Place &a, const Place &b);
|
||||
std::string DebugString() const;
|
||||
};
|
||||
|
||||
std::string DebugString() const;
|
||||
};
|
||||
|
||||
} // namespace lite_api
|
||||
} // namespace lite_api
|
||||
} // namespace paddle
|
||||
|
||||
@@ -1,781 +1,392 @@
|
||||
#pragma once
|
||||
|
||||
#include "paddle_lite_factory_helper.h"
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kHost, kAny, kNCHW, fp32);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kHost, kAny, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kHost, kAny, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kHost, kAny, kNCHW, int16);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kHost, kAny, kNCHW, uint8);
|
||||
|
||||
USE_LITE_KERNEL(mul, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(relu, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(leaky_relu, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(relu_clipped, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(prelu, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sigmoid, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(tanh, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(swish, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(relu6, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(log, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(exp, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(floor, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(hard_sigmoid, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(rsqrt, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(square, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(hard_swish, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reciprocal, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(abs, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(thresholded_relu, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elu, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(scatter_nd_add, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(scatter_nd_add, kHost, kFloat, kNCHW, float32_int64);
|
||||
|
||||
USE_LITE_KERNEL(scatter_nd_add, kHost, kFloat, kNCHW, int32_int32);
|
||||
|
||||
USE_LITE_KERNEL(scatter_nd_add, kHost, kFloat, kNCHW, int32_int64);
|
||||
|
||||
USE_LITE_KERNEL(scatter_nd_add, kHost, kFloat, kNCHW, int64_int32);
|
||||
|
||||
USE_LITE_KERNEL(scatter_nd_add, kHost, kFloat, kNCHW, int64_int64);
|
||||
|
||||
USE_LITE_KERNEL(strided_slice, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(strided_slice, kHost, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(strided_slice, kHost, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(top_k, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(split_lod_tensor, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(inverse, kHost, kFloat, kNCHW, fp32);
|
||||
|
||||
USE_LITE_KERNEL(box_coder, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(scatter, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_pad, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_pad, kHost, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(sequence_pad, kHost, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(sequence_softmax, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(conv2d, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(depthwise_conv2d, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(conv2d, kARM, kInt8, kNCHW, int8_out);
|
||||
|
||||
USE_LITE_KERNEL(conv2d, kARM, kInt8, kNCHW, fp32_out);
|
||||
|
||||
USE_LITE_KERNEL(depthwise_conv2d, kARM, kInt8, kNCHW, int8_out);
|
||||
|
||||
USE_LITE_KERNEL(depthwise_conv2d, kARM, kInt8, kNCHW, fp32_out);
|
||||
|
||||
USE_LITE_KERNEL(sum, kARM, kFloat, kNCHW, sum_i32);
|
||||
|
||||
USE_LITE_KERNEL(sum, kARM, kFloat, kNCHW, sum_i64);
|
||||
|
||||
USE_LITE_KERNEL(sum, kARM, kFloat, kNCHW, sum_fp32);
|
||||
|
||||
USE_LITE_KERNEL(gather_tree, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(gather_tree, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(increment, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(write_back, kHost, kAny, kAny, write_back);
|
||||
|
||||
USE_LITE_KERNEL(index_select, kHost, kAny, kNCHW, fp32);
|
||||
|
||||
USE_LITE_KERNEL(index_select, kHost, kAny, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(index_select, kHost, kAny, kNCHW, int16);
|
||||
|
||||
USE_LITE_KERNEL(index_select, kHost, kAny, kNCHW, int8);
|
||||
|
||||
USE_LITE_KERNEL(yolo_box, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_prod, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_prod, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_prod, kARM, kFloat, kNCHW, reduce_prod_i64);
|
||||
|
||||
USE_LITE_KERNEL(reduce_prod, kARM, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(cos, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(conditional_block, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(crf_decoding, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(density_prior_box, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fill_constant_batch_size_like, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(deformable_conv, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(roi_align, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(tensor_array_to_tensor, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(print, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_sum, kARM, kFloat, kNCHW, def_int32);
|
||||
|
||||
USE_LITE_KERNEL(reduce_sum, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(expand, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(clip, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(generate_proposals, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(select_input, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt8, kNCHW, fp32_to_int8);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt32, kNCHW, int32_to_fp32);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt32, kNCHW, int32_to_int64);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt32, kNCHW, fp32_to_int32);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt64, kNCHW, int64_to_fp32);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt8, kNCHW, int8_to_fp32);
|
||||
|
||||
USE_LITE_KERNEL(calib, kARM, kInt64, kNCHW, int64_to_int32);
|
||||
|
||||
USE_LITE_KERNEL(calib_once, kARM, kInt8, kNCHW, fp32_to_int8);
|
||||
|
||||
USE_LITE_KERNEL(calib_once, kARM, kInt8, kNCHW, int8_to_fp32);
|
||||
|
||||
USE_LITE_KERNEL(calib_once, kARM, kInt64, kNCHW, int64_to_int32);
|
||||
|
||||
USE_LITE_KERNEL(affine_channel, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fill_zeros_like, kHost, kFloat, kNCHW, float32);
|
||||
|
||||
USE_LITE_KERNEL(fill_zeros_like, kHost, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(fill_zeros_like, kHost, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(reverse, kHost, kAny, kNCHW, fp32);
|
||||
|
||||
USE_LITE_KERNEL(depthwise_conv2d_transpose, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(depthwise_conv2d_transpose, kARM, kInt8, kNCHW, fp32_out);
|
||||
|
||||
USE_LITE_KERNEL(depthwise_conv2d_transpose, kARM, kInt8, kNCHW, int8_out);
|
||||
|
||||
USE_LITE_KERNEL(decode_bboxes, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_unpad, kHost, kFloat, kAny, float32);
|
||||
|
||||
USE_LITE_KERNEL(sequence_unpad, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(shape, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(meshgrid, kHost, kFloat, kAny, float32);
|
||||
|
||||
USE_LITE_KERNEL(meshgrid, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(flatten_contiguous_range, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(pixel_shuffle, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(gru_unit, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(negative, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(axpy, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(gather, kHost, kFloat, kNCHW, int32int32);
|
||||
|
||||
USE_LITE_KERNEL(gather, kHost, kFloat, kNCHW, int64int64);
|
||||
|
||||
USE_LITE_KERNEL(gather, kHost, kFloat, kNCHW, int64int32);
|
||||
|
||||
USE_LITE_KERNEL(gather, kHost, kFloat, kNCHW, int32int64);
|
||||
|
||||
USE_LITE_KERNEL(equal, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(equal, kHost, kInt64, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(equal, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(equal, kHost, kInt32, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(equal, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(not_equal, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(not_equal, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(not_equal, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(less_than, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(less_than, kHost, kInt32, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(less_than, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(less_than, kHost, kInt64, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(less_than, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(less_equal, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(less_equal, kHost, kInt64, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(less_equal, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(greater_than, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(greater_than, kHost, kFloat, kAny, bool);
|
||||
|
||||
USE_LITE_KERNEL(greater_than, kHost, kInt32, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(greater_than, kHost, kInt64, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(greater_than, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(greater_equal, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(greater_equal, kHost, kFloat, kAny, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(multiclass_nms, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(multiclass_nms2, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(multiclass_nms3, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(gaussian_random, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(read_from_array, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(softmax, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(split, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(split, kHost, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(split, kHost, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(split, kHost, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(group_norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(instance_norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(pad2d, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(layer_norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(assign_value, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(one_hot, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(one_hot_v2, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(one_hot_v2, kHost, kAny, kAny, one_hot_v2_int32);
|
||||
|
||||
USE_LITE_KERNEL(distribute_fpn_proposals, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sin, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fc, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fc, kARM, kInt8, kNCHW, int8out);
|
||||
|
||||
USE_LITE_KERNEL(fc, kARM, kInt8, kNCHW, fp32out);
|
||||
|
||||
USE_LITE_KERNEL(unique_with_counts, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(unstack, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(unstack, kHost, kFloat, kAny, unstack_int32);
|
||||
|
||||
USE_LITE_KERNEL(collect_fpn_proposals, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(transpose, kARM, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(transpose2, kARM, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(assign, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(assign, kHost, kAny, kAny, def_tensor_array);
|
||||
|
||||
USE_LITE_KERNEL(lookup_table_dequant, kARM, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(crop_tensor, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(crop_tensor, kHost, kFloat, kAny, int32_precision);
|
||||
|
||||
USE_LITE_KERNEL(feed, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(matmul, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(matmul, kARM, kInt8, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(scale, kARM, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(scale, kARM, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(scale, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(scale, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(scale, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(dropout, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(grid_sampler, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(pow, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fetch, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(squeeze, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(squeeze2, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(affine_grid, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(crop, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(crop, kHost, kInt32, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(merge_lod_tensor, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(cos_sim, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(lod_reset, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(relu, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(leaky_relu, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(prelu, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sigmoid, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(tanh, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(relu6, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(thresholded_relu, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elu, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(bilinear_interp, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(nearest_interp, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(bilinear_interp_v2, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(nearest_interp_v2, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(correlation, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(tril_triu, kHost, kAny, kNCHW, float32);
|
||||
|
||||
USE_LITE_KERNEL(uniform_random, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(is_empty, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(anchor_generator, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(logical_xor, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(logical_and, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(logical_or, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(logical_not, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(mean, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(write_to_array, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(sparse_conv2d, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sparse_conv2d, kARM, kInt8, kNCHW, int8_fp32_out);
|
||||
|
||||
USE_LITE_KERNEL(sparse_conv2d, kARM, kInt8, kNCHW, int8_int8_out);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kARM, kAny, kNCHW, fp32);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kARM, kAny, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kARM, kAny, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kARM, kAny, kNCHW, int16);
|
||||
|
||||
USE_LITE_KERNEL(arg_max, kARM, kAny, kNCHW, uint8);
|
||||
|
||||
USE_LITE_KERNEL(stack, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(stack, kHost, kFloat, kAny, int32_def);
|
||||
|
||||
USE_LITE_KERNEL(stack, kHost, kFloat, kAny, int64_def);
|
||||
|
||||
USE_LITE_KERNEL(where_index, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_mean, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(beam_search_decode, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(argsort, kHost, kFloat, kAny, argsort_fp32);
|
||||
|
||||
USE_LITE_KERNEL(argsort, kHost, kFloat, kAny, argsort_int32);
|
||||
|
||||
USE_LITE_KERNEL(argsort, kHost, kFloat, kAny, argsort_int64);
|
||||
|
||||
USE_LITE_KERNEL(lookup_table, kARM, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(lookup_table_v2, kARM, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(beam_search, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(pool2d, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(slice, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(slice, kARM, kFloat, kNCHW, float_i64_starts_ends);
|
||||
|
||||
USE_LITE_KERNEL(slice, kARM, kFloat, kNCHW, bool_slice);
|
||||
|
||||
USE_LITE_KERNEL(slice, kARM, kFloat, kNCHW, int32_slice);
|
||||
|
||||
USE_LITE_KERNEL(slice, kARM, kFloat, kNCHW, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(expand_as, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(expand_as, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(batch_norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sync_batch_norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(ctc_align, kHost, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(ctc_align, kHost, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(where, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(conv2d_transpose, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(conv2d_transpose, kARM, kInt8, kNCHW, fp32_out);
|
||||
|
||||
USE_LITE_KERNEL(conv2d_transpose, kARM, kInt8, kNCHW, int8_out);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_add, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_add, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_add, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_add, kARM, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_add, kARM, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_add_activation, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_sub, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_sub, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_sub, kARM, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_sub, kARM, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_sub_activation, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mul, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mul, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mul, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mul, kARM, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mul, kARM, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_mul_activation, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_mul_activation, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_max, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_max_activation, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_min, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_min_activation, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_div, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_div, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_div, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fusion_elementwise_div_activation, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mod, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mod, kARM, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_mod, kARM, kFloat, kNCHW, int32_mod);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_pow, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_pow, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_floordiv, kARM, kInt32, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_floordiv, kARM, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_floordiv, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(elementwise_floordiv, kARM, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(rnn, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_min, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_min, kARM, kFloat, kNCHW, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(sequence_expand, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(lstm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(lstm, kARM, kInt8, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(pixel_shuffle, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(box_coder, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fill_any_like, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(fill_zeros_like, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(generate_proposals_v2, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(pad2d, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(matmul_v2, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(retinanet_detection_output, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(shuffle_channel, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(cast, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(lrn, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(prior_box, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(unsqueeze, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(unsqueeze2, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(fill_constant, kHost, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(box_clip, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(unbind, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(unbind, kHost, kInt64, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(p_norm, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_expand_as, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(tile, kHost, kFloat, kNCHW, def_float);
|
||||
|
||||
USE_LITE_KERNEL(tile, kHost, kInt32, kNCHW, def_int32);
|
||||
|
||||
USE_LITE_KERNEL(tile, kHost, kInt64, kNCHW, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(tile, kHost, kFloat, kNCHW, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(tile, kHost, kInt8, kNCHW, def_int8);
|
||||
|
||||
USE_LITE_KERNEL(tile, kHost, kFloat, kNCHW, def_bool);
|
||||
|
||||
USE_LITE_KERNEL(norm, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(p_norm, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_max, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_max, kARM, kFloat, kNCHW, i64);
|
||||
|
||||
USE_LITE_KERNEL(gru, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(gru, kARM, kInt8, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_all, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reduce_any, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reshape, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(reshape2, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(flatten, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(flatten2, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(concat, kARM, kAny, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_conv, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(relu_clipped, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(swish, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(log, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(exp, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(floor, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(hard_sigmoid, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sqrt, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(rsqrt, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(square, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(hard_swish, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(reciprocal, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(abs, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(gelu, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(erf, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sign, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(softplus, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(mish, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(lod_array_length, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(deformable_conv, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(layout, kARM, kFloat, kNCHW, nchw2nhwc);
|
||||
|
||||
USE_LITE_KERNEL(layout, kARM, kFloat, kNCHW, nhwc2nchw);
|
||||
|
||||
USE_LITE_KERNEL(layout, kARM, kInt8, kNCHW, int8_nchw2nhwc);
|
||||
|
||||
USE_LITE_KERNEL(layout, kARM, kInt8, kNCHW, int8_nhwc2nchw);
|
||||
|
||||
USE_LITE_KERNEL(layout_once, kARM, kFloat, kNCHW, nchw2nhwc);
|
||||
|
||||
USE_LITE_KERNEL(layout_once, kARM, kFloat, kNCHW, nhwc2nchw);
|
||||
|
||||
USE_LITE_KERNEL(layout_once, kARM, kInt8, kNCHW, int8_nchw2nhwc);
|
||||
|
||||
USE_LITE_KERNEL(layout_once, kARM, kInt8, kNCHW, int8_nhwc2nchw);
|
||||
|
||||
USE_LITE_KERNEL(roi_perspective_transform, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(while, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(flip, kHost, kAny, kNCHW, flip_fp32);
|
||||
|
||||
USE_LITE_KERNEL(flip, kHost, kAny, kNCHW, flip_i64);
|
||||
|
||||
USE_LITE_KERNEL(top_k_v2, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(pad3d, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(range, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(range, kHost, kInt64, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(range, kHost, kInt32, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(range, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(range, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(im2sequence, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(unfold, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(unfold, kHost, kFloat, kNCHW, def_int32);
|
||||
|
||||
USE_LITE_KERNEL(unfold, kHost, kFloat, kNCHW, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(unfold, kHost, kInt8, kNCHW, def_int8);
|
||||
|
||||
USE_LITE_KERNEL(gather_nd, kHost, kAny, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(polygon_box_transform, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(expand_v2, kHost, kFloat, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(expand_v2, kHost, kInt32, kAny, def);
|
||||
|
||||
USE_LITE_KERNEL(expand_v2, kHost, kFloat, kAny, def_int64);
|
||||
|
||||
USE_LITE_KERNEL(sequence_pool, kARM, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_mask, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(sequence_mask, kHost, kFloat, kNCHW, int32);
|
||||
|
||||
USE_LITE_KERNEL(sequence_mask, kHost, kFloat, kNCHW, int64);
|
||||
|
||||
USE_LITE_KERNEL(cumsum, kHost, kFloat, kAny, float32);
|
||||
|
||||
USE_LITE_KERNEL(cumsum, kHost, kFloat, kAny, int32);
|
||||
|
||||
USE_LITE_KERNEL(cumsum, kHost, kFloat, kAny, int64);
|
||||
|
||||
USE_LITE_KERNEL(matrix_nms, kHost, kFloat, kNCHW, def);
|
||||
|
||||
USE_LITE_KERNEL(linspace, kHost, kFloat, kAny, float32);
|
||||
|
||||
USE_LITE_KERNEL(linspace, kHost, kInt32, kAny, int32);
|
||||
@@ -1,547 +1,275 @@
|
||||
#pragma once
|
||||
|
||||
#include "paddle_lite_factory_helper.h"
|
||||
|
||||
USE_LITE_OP(tensor_array_to_tensor);
|
||||
|
||||
USE_LITE_OP(beam_search);
|
||||
|
||||
USE_LITE_OP(index_select);
|
||||
|
||||
USE_LITE_OP(fusion_elementwise_sub_activation);
|
||||
|
||||
USE_LITE_OP(fusion_elementwise_add_activation);
|
||||
|
||||
USE_LITE_OP(fusion_elementwise_mul_activation);
|
||||
|
||||
USE_LITE_OP(fusion_elementwise_max_activation);
|
||||
|
||||
USE_LITE_OP(fusion_elementwise_min_activation);
|
||||
|
||||
USE_LITE_OP(fusion_elementwise_div_activation);
|
||||
|
||||
USE_LITE_OP(softmax);
|
||||
|
||||
USE_LITE_OP(io_copy);
|
||||
|
||||
USE_LITE_OP(fake_quantize_moving_average_abs_max);
|
||||
|
||||
USE_LITE_OP(scatter_nd_add);
|
||||
|
||||
USE_LITE_OP(sum);
|
||||
|
||||
USE_LITE_OP(norm);
|
||||
|
||||
USE_LITE_OP(p_norm);
|
||||
|
||||
USE_LITE_OP(gather_tree);
|
||||
|
||||
USE_LITE_OP(collect_fpn_proposals);
|
||||
|
||||
USE_LITE_OP(topk_pooling);
|
||||
|
||||
USE_LITE_OP(axpy);
|
||||
|
||||
USE_LITE_OP(expand_v2);
|
||||
|
||||
USE_LITE_OP(top_k_v2);
|
||||
|
||||
USE_LITE_OP(generate_proposals);
|
||||
|
||||
USE_LITE_OP(write_back);
|
||||
|
||||
USE_LITE_OP(conv2d);
|
||||
|
||||
USE_LITE_OP(conv3d);
|
||||
|
||||
USE_LITE_OP(depthwise_conv2d);
|
||||
|
||||
USE_LITE_OP(assign);
|
||||
|
||||
USE_LITE_OP(sequence_reverse_embedding);
|
||||
|
||||
USE_LITE_OP(cast);
|
||||
|
||||
USE_LITE_OP(lstm);
|
||||
|
||||
USE_LITE_OP(subgraph);
|
||||
|
||||
USE_LITE_OP(gru);
|
||||
|
||||
USE_LITE_OP(box_coder);
|
||||
|
||||
USE_LITE_OP(sequence_pool_concat);
|
||||
|
||||
USE_LITE_OP(crop);
|
||||
|
||||
USE_LITE_OP(print);
|
||||
|
||||
USE_LITE_OP(one_hot_v2);
|
||||
|
||||
USE_LITE_OP(sequence_reshape);
|
||||
|
||||
USE_LITE_OP(fake_quantize_range_abs_max);
|
||||
|
||||
USE_LITE_OP(fake_quantize_abs_max);
|
||||
|
||||
USE_LITE_OP(fill_zeros_like);
|
||||
|
||||
USE_LITE_OP(multiclass_nms);
|
||||
|
||||
USE_LITE_OP(multiclass_nms2);
|
||||
|
||||
USE_LITE_OP(multiclass_nms3);
|
||||
|
||||
USE_LITE_OP(tril_triu);
|
||||
|
||||
USE_LITE_OP(argsort);
|
||||
|
||||
USE_LITE_OP(fill_constant_batch_size_like);
|
||||
|
||||
USE_LITE_OP(cos);
|
||||
|
||||
USE_LITE_OP(matrix_nms);
|
||||
|
||||
USE_LITE_OP(lrn);
|
||||
|
||||
USE_LITE_OP(arg_max);
|
||||
|
||||
USE_LITE_OP(decode_bboxes);
|
||||
|
||||
USE_LITE_OP(layout_once);
|
||||
|
||||
USE_LITE_OP(equal);
|
||||
|
||||
USE_LITE_OP(not_equal);
|
||||
|
||||
USE_LITE_OP(less_than);
|
||||
|
||||
USE_LITE_OP(less_equal);
|
||||
|
||||
USE_LITE_OP(greater_than);
|
||||
|
||||
USE_LITE_OP(greater_equal);
|
||||
|
||||
USE_LITE_OP(sequence_mask);
|
||||
|
||||
USE_LITE_OP(matmul_v2);
|
||||
|
||||
USE_LITE_OP(search_seq_softmax);
|
||||
|
||||
USE_LITE_OP(pad2d);
|
||||
|
||||
USE_LITE_OP(search_grnn);
|
||||
|
||||
USE_LITE_OP(flip);
|
||||
|
||||
USE_LITE_OP(expand_as);
|
||||
|
||||
USE_LITE_OP(generate_proposals_v2);
|
||||
|
||||
USE_LITE_OP(distribute_fpn_proposals);
|
||||
|
||||
USE_LITE_OP(rnn);
|
||||
|
||||
USE_LITE_OP(fake_channel_wise_dequantize_max_abs);
|
||||
|
||||
USE_LITE_OP(search_aligned_mat_mul);
|
||||
|
||||
USE_LITE_OP(concat);
|
||||
|
||||
USE_LITE_OP(attention_padding_mask);
|
||||
|
||||
USE_LITE_OP(search_attention_padding_mask);
|
||||
|
||||
USE_LITE_OP(retinanet_detection_output);
|
||||
|
||||
USE_LITE_OP(fpga_conv2d);
|
||||
|
||||
USE_LITE_OP(pad3d);
|
||||
|
||||
USE_LITE_OP(reverse);
|
||||
|
||||
USE_LITE_OP(layout);
|
||||
|
||||
USE_LITE_OP(range);
|
||||
|
||||
USE_LITE_OP(split);
|
||||
|
||||
USE_LITE_OP(__xpu__dynamic_lstm_fuse_op);
|
||||
|
||||
USE_LITE_OP(stack);
|
||||
|
||||
USE_LITE_OP(__xpu__squeeze_excitation_block);
|
||||
|
||||
USE_LITE_OP(mean);
|
||||
|
||||
USE_LITE_OP(prior_box);
|
||||
|
||||
USE_LITE_OP(search_fc);
|
||||
|
||||
USE_LITE_OP(__xpu__generate_sequence);
|
||||
|
||||
USE_LITE_OP(elementwise_sub);
|
||||
|
||||
USE_LITE_OP(elementwise_add);
|
||||
|
||||
USE_LITE_OP(elementwise_mul);
|
||||
|
||||
USE_LITE_OP(elementwise_max);
|
||||
|
||||
USE_LITE_OP(elementwise_min);
|
||||
|
||||
USE_LITE_OP(elementwise_div);
|
||||
|
||||
USE_LITE_OP(elementwise_floordiv);
|
||||
|
||||
USE_LITE_OP(elementwise_mod);
|
||||
|
||||
USE_LITE_OP(elementwise_pow);
|
||||
|
||||
USE_LITE_OP(sequence_pool);
|
||||
|
||||
USE_LITE_OP(correlation);
|
||||
|
||||
USE_LITE_OP(sequence_unpad);
|
||||
|
||||
USE_LITE_OP(cumsum);
|
||||
|
||||
USE_LITE_OP(lod_array_length);
|
||||
|
||||
USE_LITE_OP(increment);
|
||||
|
||||
USE_LITE_OP(__xpu__multi_softmax);
|
||||
|
||||
USE_LITE_OP(fill_constant);
|
||||
|
||||
USE_LITE_OP(group_norm);
|
||||
|
||||
USE_LITE_OP(scatter);
|
||||
|
||||
USE_LITE_OP(lod_reset);
|
||||
|
||||
USE_LITE_OP(shuffle_channel);
|
||||
|
||||
USE_LITE_OP(gather_nd);
|
||||
|
||||
USE_LITE_OP(search_seq_depadding);
|
||||
|
||||
USE_LITE_OP(cos_sim);
|
||||
|
||||
USE_LITE_OP(is_empty);
|
||||
|
||||
USE_LITE_OP(fill_any_like);
|
||||
|
||||
USE_LITE_OP(match_matrix_tensor);
|
||||
|
||||
USE_LITE_OP(yolo_box);
|
||||
|
||||
USE_LITE_OP(__xpu__fc);
|
||||
|
||||
USE_LITE_OP(where);
|
||||
|
||||
USE_LITE_OP(fake_quantize_dequantize_abs_max);
|
||||
|
||||
USE_LITE_OP(grid_sampler);
|
||||
|
||||
USE_LITE_OP(conv2d_transpose);
|
||||
|
||||
USE_LITE_OP(depthwise_conv2d_transpose);
|
||||
|
||||
USE_LITE_OP(__xpu__bigru);
|
||||
|
||||
USE_LITE_OP(calib_once);
|
||||
|
||||
USE_LITE_OP(sequence_conv);
|
||||
|
||||
USE_LITE_OP(__xpu__resnet_cbam);
|
||||
|
||||
USE_LITE_OP(pixel_shuffle);
|
||||
|
||||
USE_LITE_OP(__xpu__logit);
|
||||
|
||||
USE_LITE_OP(sin);
|
||||
|
||||
USE_LITE_OP(beam_search_decode);
|
||||
|
||||
USE_LITE_OP(polygon_box_transform);
|
||||
|
||||
USE_LITE_OP(split_lod_tensor);
|
||||
|
||||
USE_LITE_OP(fc);
|
||||
|
||||
USE_LITE_OP(var_conv_2d);
|
||||
|
||||
USE_LITE_OP(calib);
|
||||
|
||||
USE_LITE_OP(sequence_concat);
|
||||
|
||||
USE_LITE_OP(__xpu__multi_encoder);
|
||||
|
||||
USE_LITE_OP(sequence_pad);
|
||||
|
||||
USE_LITE_OP(crf_decoding);
|
||||
|
||||
USE_LITE_OP(reshape);
|
||||
|
||||
USE_LITE_OP(reshape2);
|
||||
|
||||
USE_LITE_OP(gaussian_random);
|
||||
|
||||
USE_LITE_OP(pool2d);
|
||||
|
||||
USE_LITE_OP(sequence_arithmetic);
|
||||
|
||||
USE_LITE_OP(search_seq_arithmetic);
|
||||
|
||||
USE_LITE_OP(transpose);
|
||||
|
||||
USE_LITE_OP(transpose2);
|
||||
|
||||
USE_LITE_OP(one_hot);
|
||||
|
||||
USE_LITE_OP(matmul);
|
||||
|
||||
USE_LITE_OP(lookup_table_v2);
|
||||
|
||||
USE_LITE_OP(unique_with_counts);
|
||||
|
||||
USE_LITE_OP(assign_value);
|
||||
|
||||
USE_LITE_OP(lookup_table_dequant);
|
||||
|
||||
USE_LITE_OP(negative);
|
||||
|
||||
USE_LITE_OP(reduce_sum);
|
||||
|
||||
USE_LITE_OP(reduce_prod);
|
||||
|
||||
USE_LITE_OP(reduce_max);
|
||||
|
||||
USE_LITE_OP(reduce_min);
|
||||
|
||||
USE_LITE_OP(reduce_all);
|
||||
|
||||
USE_LITE_OP(reduce_any);
|
||||
|
||||
USE_LITE_OP(reduce_mean);
|
||||
|
||||
USE_LITE_OP(pow);
|
||||
|
||||
USE_LITE_OP(__xpu__conv2d);
|
||||
|
||||
USE_LITE_OP(uniform_random);
|
||||
|
||||
USE_LITE_OP(nearest_interp);
|
||||
|
||||
USE_LITE_OP(bilinear_interp);
|
||||
|
||||
USE_LITE_OP(roi_align);
|
||||
|
||||
USE_LITE_OP(dropout);
|
||||
|
||||
USE_LITE_OP(write_to_array);
|
||||
|
||||
USE_LITE_OP(bilinear_interp_v2);
|
||||
|
||||
USE_LITE_OP(nearest_interp_v2);
|
||||
|
||||
USE_LITE_OP(strided_slice);
|
||||
|
||||
USE_LITE_OP(feed);
|
||||
|
||||
USE_LITE_OP(fake_quantize_dequantize_moving_average_abs_max);
|
||||
|
||||
USE_LITE_OP(ctc_align);
|
||||
|
||||
USE_LITE_OP(scale);
|
||||
|
||||
USE_LITE_OP(linspace);
|
||||
|
||||
USE_LITE_OP(sequence_topk_avg_pooling);
|
||||
|
||||
USE_LITE_OP(im2sequence);
|
||||
|
||||
USE_LITE_OP(density_prior_box);
|
||||
|
||||
USE_LITE_OP(roi_perspective_transform);
|
||||
|
||||
USE_LITE_OP(while);
|
||||
|
||||
USE_LITE_OP(top_k);
|
||||
|
||||
USE_LITE_OP(mul);
|
||||
|
||||
USE_LITE_OP(expand);
|
||||
|
||||
USE_LITE_OP(__xpu__sfa_head);
|
||||
|
||||
USE_LITE_OP(sequence_expand);
|
||||
|
||||
USE_LITE_OP(where_index);
|
||||
|
||||
USE_LITE_OP(flatten);
|
||||
|
||||
USE_LITE_OP(flatten2);
|
||||
|
||||
USE_LITE_OP(flatten_contiguous_range);
|
||||
|
||||
USE_LITE_OP(max_pool2d_with_index);
|
||||
|
||||
USE_LITE_OP(unstack);
|
||||
|
||||
USE_LITE_OP(unsqueeze);
|
||||
|
||||
USE_LITE_OP(unsqueeze2);
|
||||
|
||||
USE_LITE_OP(affine_grid);
|
||||
|
||||
USE_LITE_OP(deformable_conv);
|
||||
|
||||
USE_LITE_OP(box_clip);
|
||||
|
||||
USE_LITE_OP(shape);
|
||||
|
||||
USE_LITE_OP(meshgrid);
|
||||
|
||||
USE_LITE_OP(__xpu__softmax_topk);
|
||||
|
||||
USE_LITE_OP(sequence_softmax);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_search_attention);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_search_attention2);
|
||||
|
||||
USE_LITE_OP(slice);
|
||||
|
||||
USE_LITE_OP(sign);
|
||||
|
||||
USE_LITE_OP(instance_norm);
|
||||
|
||||
USE_LITE_OP(io_copy_once);
|
||||
|
||||
USE_LITE_OP(gather);
|
||||
|
||||
USE_LITE_OP(__xpu__embedding_with_eltwise_add);
|
||||
|
||||
USE_LITE_OP(inverse);
|
||||
|
||||
USE_LITE_OP(__xpu__resnet50);
|
||||
|
||||
USE_LITE_OP(unfold);
|
||||
|
||||
USE_LITE_OP(search_seq_fc);
|
||||
|
||||
USE_LITE_OP(sequence_reverse);
|
||||
|
||||
USE_LITE_OP(sparse_conv2d);
|
||||
|
||||
USE_LITE_OP(layer_norm);
|
||||
|
||||
USE_LITE_OP(read_from_array);
|
||||
|
||||
USE_LITE_OP(fake_channel_wise_quantize_dequantize_abs_max);
|
||||
|
||||
USE_LITE_OP(search_group_padding);
|
||||
|
||||
USE_LITE_OP(select_input);
|
||||
|
||||
USE_LITE_OP(affine_channel);
|
||||
|
||||
USE_LITE_OP(clip);
|
||||
|
||||
USE_LITE_OP(fetch);
|
||||
|
||||
USE_LITE_OP(anchor_generator);
|
||||
|
||||
USE_LITE_OP(sequence_expand_as);
|
||||
|
||||
USE_LITE_OP(lookup_table);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_bid_emb_grnn_att);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_bid_emb_grnn_att2);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_bid_emb_att);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_match_conv_topk);
|
||||
|
||||
USE_LITE_OP(__xpu__mmdnn_merge_all);
|
||||
|
||||
USE_LITE_OP(squeeze);
|
||||
|
||||
USE_LITE_OP(squeeze2);
|
||||
|
||||
USE_LITE_OP(tile);
|
||||
|
||||
USE_LITE_OP(merge_lod_tensor);
|
||||
|
||||
USE_LITE_OP(conditional_block);
|
||||
|
||||
USE_LITE_OP(logical_xor);
|
||||
|
||||
USE_LITE_OP(logical_and);
|
||||
|
||||
USE_LITE_OP(logical_or);
|
||||
|
||||
USE_LITE_OP(logical_not);
|
||||
|
||||
USE_LITE_OP(crop_tensor);
|
||||
|
||||
USE_LITE_OP(fake_dequantize_max_abs);
|
||||
|
||||
USE_LITE_OP(batch_norm);
|
||||
|
||||
USE_LITE_OP(sync_batch_norm);
|
||||
|
||||
USE_LITE_OP(gru_unit);
|
||||
|
||||
USE_LITE_OP(square);
|
||||
|
||||
USE_LITE_OP(relu_clipped);
|
||||
|
||||
USE_LITE_OP(swish);
|
||||
|
||||
USE_LITE_OP(log);
|
||||
|
||||
USE_LITE_OP(exp);
|
||||
|
||||
USE_LITE_OP(abs);
|
||||
|
||||
USE_LITE_OP(floor);
|
||||
|
||||
USE_LITE_OP(hard_sigmoid);
|
||||
|
||||
USE_LITE_OP(sqrt);
|
||||
|
||||
USE_LITE_OP(rsqrt);
|
||||
|
||||
USE_LITE_OP(softsign);
|
||||
|
||||
USE_LITE_OP(gelu);
|
||||
|
||||
USE_LITE_OP(hard_swish);
|
||||
|
||||
USE_LITE_OP(reciprocal);
|
||||
|
||||
USE_LITE_OP(mish);
|
||||
|
||||
USE_LITE_OP(sigmoid);
|
||||
|
||||
USE_LITE_OP(tanh);
|
||||
|
||||
USE_LITE_OP(relu);
|
||||
|
||||
USE_LITE_OP(leaky_relu);
|
||||
|
||||
USE_LITE_OP(relu6);
|
||||
|
||||
USE_LITE_OP(prelu);
|
||||
|
||||
USE_LITE_OP(thresholded_relu);
|
||||
|
||||
USE_LITE_OP(elu);
|
||||
|
||||
USE_LITE_OP(erf);
|
||||
|
||||
USE_LITE_OP(softplus);
|
||||
@@ -13,207 +13,108 @@
|
||||
// limitations under the License.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "paddle_lite_factory_helper.h" // NOLINT
|
||||
|
||||
USE_MIR_PASS(demo);
|
||||
|
||||
USE_MIR_PASS(static_kernel_pick_pass);
|
||||
|
||||
USE_MIR_PASS(op_transformation_pass);
|
||||
|
||||
USE_MIR_PASS(variable_place_inference_pass);
|
||||
|
||||
USE_MIR_PASS(type_target_cast_pass);
|
||||
|
||||
USE_MIR_PASS(lite_unsqueeze2_pad3d_squeeze2_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__fpga_kernel_place_correct_pass);
|
||||
|
||||
USE_MIR_PASS(opencl_kernel_place_correct_pass);
|
||||
|
||||
USE_MIR_PASS(generate_program_pass);
|
||||
|
||||
USE_MIR_PASS(io_copy_kernel_pick_pass);
|
||||
|
||||
USE_MIR_PASS(argument_type_display_pass);
|
||||
|
||||
USE_MIR_PASS(runtime_context_assign_pass);
|
||||
|
||||
USE_MIR_PASS(graph_visualize_pass);
|
||||
|
||||
USE_MIR_PASS(sparse_conv_detect_pass)
|
||||
|
||||
USE_MIR_PASS(adaptive_1x1_pool2d_convert_global_pass);
|
||||
|
||||
USE_MIR_PASS(remove_scale1_pass);
|
||||
|
||||
USE_MIR_PASS(remove_tf_redundant_ops_pass);
|
||||
|
||||
USE_MIR_PASS(lite_conv_bn_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_conv_conv_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_squeeze2_matmul_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_reshape2_matmul_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_matmul_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_fc_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_matmul_element_add_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_shuffle_channel_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_transpose_softmax_transpose_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_interpolate_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_sequence_pool_concat_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(identity_scale_eliminate_pass);
|
||||
|
||||
USE_MIR_PASS(identity_dropout_eliminate_pass);
|
||||
|
||||
USE_MIR_PASS(lite_conv_elementwise_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_conv_activation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_var_conv_2d_activation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_match_matrix_activation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_scales_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_scaleacts_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_sequence_reverse_embedding_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_elementwise_activation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_elementwise_scale_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_conv_scale_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_conv_elementwise_tree_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_quant_dequant_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(type_precision_cast_pass);
|
||||
|
||||
USE_MIR_PASS(type_layout_cast_pass);
|
||||
|
||||
USE_MIR_PASS(type_layout_cast_preprocess_pass);
|
||||
|
||||
USE_MIR_PASS(memory_optimize_pass);
|
||||
|
||||
USE_MIR_PASS(xpu_memory_optimize_pass);
|
||||
|
||||
USE_MIR_PASS(lite_inplace_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(multi_stream_analysis_pass);
|
||||
|
||||
USE_MIR_PASS(elementwise_mul_constant_eliminate_pass)
|
||||
|
||||
USE_MIR_PASS(npu_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(huawei_ascend_npu_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(imagination_nna_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(nnadapter_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(xpu_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(mlu_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(mlu_postprocess_pass);
|
||||
|
||||
USE_MIR_PASS(weight_quantization_preprocess_pass);
|
||||
|
||||
USE_MIR_PASS(post_quant_dynamic_pass);
|
||||
|
||||
USE_MIR_PASS(fp16_attribute_pass);
|
||||
|
||||
USE_MIR_PASS(apu_subgraph_pass);
|
||||
|
||||
USE_MIR_PASS(fpga_concat_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(quantized_op_attributes_inference_pass);
|
||||
|
||||
USE_MIR_PASS(quantization_parameters_propagation_pass);
|
||||
|
||||
USE_MIR_PASS(restrict_quantized_op_with_same_input_output_scale_pass);
|
||||
|
||||
USE_MIR_PASS(control_flow_op_unused_inputs_and_outputs_eliminate_pass);
|
||||
|
||||
USE_MIR_PASS(control_flow_op_shared_inputs_and_outputs_place_sync_pass);
|
||||
|
||||
USE_MIR_PASS(lite_scale_activation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_instance_norm_activation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(ssd_boxes_calc_offline_pass);
|
||||
|
||||
USE_MIR_PASS(fix_mismatched_precision_pass);
|
||||
|
||||
USE_MIR_PASS(lite_flatten_fc_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_fc_prelu_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(lite_greater_than_cast_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(assign_value_calc_offline_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__graph_dedup_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__resnet_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__resnet_cbam_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__multi_encoder_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__embedding_with_eltwise_add_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__fc_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__mmdnn_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__conv2d_affine_channel_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__conv2d_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__sfa_head_meanstd_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__sfa_head_moment_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__softmax_topk_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__multi_encoder_adaptive_seqlen_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__multi_encoder_slice_link_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__generate_sequence_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__logit_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__link_previous_out_max_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__squeeze_excitation_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__bigru_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__dynamic_lstm_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__multi_softmax_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(__xpu__max_pooling_pad_zero_detect_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(x86_int8_attribute_pass);
|
||||
|
||||
USE_MIR_PASS(fill_range_fuse_pass);
|
||||
|
||||
USE_MIR_PASS(range_calc_offline_pass);
|
||||
|
||||
USE_MIR_PASS(p_norm_fill_constant_max_div_fuse_pass);
|
||||
|
||||
Reference in New Issue
Block a user