morapi_Dnn Struct Reference
Deep neural network.
More...
#include <morapi_softneuro.h>
Public Member Functions
Primary functions
Optional settings
Tensor operations
| morapi_Tensor * | morapi_Dnn_addTensor (morapi_Dnn *self, const morapi_Char *i_name) |
| morapi_Result | morapi_Dnn_removeTensor (morapi_Dnn *self, morapi_Tensor *io_tensor) |
| morapi_Int32 | morapi_Dnn_getTensorNum (const morapi_Dnn *self) |
| morapi_Tensor * | morapi_Dnn_getTensor (const morapi_Dnn *self, morapi_Int32 i_index) |
| morapi_Tensor * | morapi_Dnn_findTensor (const morapi_Dnn *self, const morapi_Char *i_name) |
Net operations
| morapi_Int32 | morapi_Dnn_getNetNum (const morapi_Dnn *self) |
| morapi_DnnNet * | morapi_Dnn_getNet (const morapi_Dnn *self, morapi_Int32 i_index) |
| morapi_DnnNet * | morapi_Dnn_findMainNet (const morapi_Dnn *self) |
| morapi_DnnNet * | morapi_Dnn_addNet (morapi_Dnn *self, const morapi_Char *i_name) |
| morapi_Result | morapi_Dnn_removeNet (morapi_Dnn *self, morapi_DnnNet *io_net) |
| morapi_DnnNet * | morapi_Dnn_findNet (const morapi_Dnn *self, const morapi_Char *i_name) |
Other functions
Related Functions
| morapi_Result | morapi_Dnn_static_installRoutineSchema (morapi_Env *env, const morapi_Char *i_schema) |
| morapi_Int32 | morapi_Dnn_static_getRoutineSchemaNum (morapi_Env *env) |
| const morapi_Char ** | morapi_Dnn_static_getRoutineSchemas (morapi_Env *env) |
Detailed Description
Deep neural network.
This structure contains all data needed to operate a dnn.
test_decompose_api.c, test_dnn_api.c, and test_prof_api.c.
Member Function Documentation
morapi_create_Dnn()
morapi_Dnn * morapi_create_Dnn (morapi_Env * io_env)
Creates a Dnn module.
Parameters
- io_env : the environment data
Returns
the created dnn module
test_decompose_api.c, test_dnn_api.c, and test_prof_api.c.
morapi_Dnn_initialize()
morapi_Result morapi_Dnn_initialize (morapi_Dnn * self)
Initializes the Dnn module.
Parameters
- self : the dnn.
morapi_Dnn_load()
morapi_Result morapi_Dnn_load (morapi_Dnn * self,
const morapi_Char * i_filename,
const void * i_pass,
morapi_Int32 i_pass_size,
morapi_Bool i_is_lazy)
Reads the dnn from a file.
Parameters
- self : the dnn module
- i_filename : the path of the dnn file
- i_pass : the password of the dnn file, use NULL for the dnn file without password
- i_pass_size : the size of the password
- i_is_lazy : lazy loading or not.
test_decompose_api.c, test_dnn_api.c, and test_prof_api.c.
morapi_Dnn_getInputNum()
morapi_Int32 morapi_Dnn_getInputNum (const morapi_Dnn * self)
Get the number of inputs for Dnn.
Parameters
- self : the dnn module
morapi_Dnn_getInputBlob()
morapi_Tensor * morapi_Dnn_getInputBlob (const morapi_Dnn * self,
morapi_Int32 i_input_index)
Gets the i_input_index-th input blob.
Valid after compiled.
Parameters
- self : the dnn module
- i_input_index : input index
morapi_Dnn_setInputImage()
morapi_Result morapi_Dnn_setInputImage (morapi_Dnn * self,
const morapi_Image * i_image,
morapi_Int32 i_index,
morapi_Int32 i_batch)
Sets an image to the i_batch-th batch of i_input_index-th input.
Uses the dnn input attrs (keep_ar, padding_color, resize_mode) for the resize settings, if available. Defaults to Bilinear resize and doesn't keep image aspect ratio (no padding).
Parameters
- self : the dnn module
- i_image : image to be set as input
- i_index : input index to set the image at
- i_batch : batch to set the image at
morapi_Dnn_setInputImageWithOptions()
morapi_Result morapi_Dnn_setInputImageWithOptions (morapi_Dnn * self,
const morapi_Image * i_image,
morapi_Int32 i_index,
morapi_Int32 i_batch,
morapi_Bool i_keep_ar,
morapi_DnnResizeMode i_resize_mode,
morapi_Uint8 i_padding_color[3])
Sets an image to the i_batch-th batch of i_input_index-th input using the given resize mode and padding to keep aspect ratio if set.
Parameters
- self : the dnn module
- i_image : image to be set as input
- i_index : input index to set the image at
- i_batch : batch to set the image at
- i_keep_ar : whether or not to keep aspect ratio if resizing the image
- i_resize_mode : image resize mode, either MORAPI_DNN_RESIZE_MODE_BILINEAR or MORAPI_DNN_RESIZE_MODE_NEAREST
- i_padding_color : 0-255 RGB values to use as padding when resizing with keep aspect ratio on, NULL for zero padding
morapi_Dnn_getInputAttrs()
morapi_Params * morapi_Dnn_getInputAttrs (const morapi_Dnn * self,
morapi_Int32 i_input_index)
Gets the i_input_index-th input attributes.
Parameters
- self : the dnn module
- i_input_index : input index
morapi_Dnn_forward()
morapi_Result morapi_Dnn_forward (morapi_Dnn * self)
Executes forward propagation once on the network.
Parameters
- self : the dnn module
morapi_Dnn_getOutputNum()
morapi_Int32 morapi_Dnn_getOutputNum (const morapi_Dnn * self)
Gets the number of outputs.
Parameters
- self : the dnn module
morapi_Dnn_getOutputAttrs()
morapi_Params * morapi_Dnn_getOutputAttrs (const morapi_Dnn * self,
morapi_Int32 i_output_index)
Gets attributes from the i_output_index-th output.
Parameters
- self : the dnn module
- i_output_index : output tensor index
Returns
morapi_Params object
morapi_Dnn_getOutputBlob()
morapi_Tensor * morapi_Dnn_getOutputBlob (const morapi_Dnn * self,
morapi_Int32 i_output_index)
Gets tensor blob from the i_output_index-th output.
Valid after compiled.
Returns
morapi_Tensor object
morapi_Dnn_save()
morapi_Result morapi_Dnn_save (morapi_Dnn * self,
const morapi_Char * o_filename,
const void * i_pass,
morapi_Int32 i_pass_size,
morapi_Bool i_is_secret)
Writes the dnn to a file.
It does nothing and returns MORAPI_ERROR_STATE if self is secret.
Parameters
- self : the dnn module
- o_filename : output file name
- i_pass : encryption password, if needed
- i_pass_size : encryption password size
- i_is_secret : whether makes the dnn secret or not.
test_decompose_api.c, and test_prof_api.c.
morapi_destroy_Dnn()
morapi_Result morapi_destroy_Dnn (morapi_Env * io_env,
morapi_Dnn * self)
Destroys the Dnn module to free memory.
Parameters
- io_env : the envitonment data
- self : the dnn module to be destroyed
test_decompose_api.c, test_dnn_api.c, and test_prof_api.c.
morapi_Dnn_setThreadNum()
morapi_Result morapi_Dnn_setThreadNum (morapi_Dnn * self,
morapi_Int32 i_thread_num)
Sets the number of threads for forward propagation.
It clears affinity masks set by morapi_Dnn_setAffinityMask(). If i_thread_num=0, the number of threads gets equal to the number of logical CPU cores.
Parameters
- self : the dnn module
- i_thread_num : number of threads to be used
morapi_Dnn_getThreadNum()
morapi_Int32 morapi_Dnn_getThreadNum (const morapi_Dnn * self)
Gets the number of CPU threads for forward propagation.
If 0 is set by morapi_Dnn_setThreadNum(), returns the number of logical CPU cores.
Parameters
- self : the dnn module
morapi_Dnn_setAffinityMask()
morapi_Result morapi_Dnn_setAffinityMask (morapi_Dnn * self,
morapi_Int32 i_thread_index,
const morapi_Char * i_affinity_mask)
Sets an affinity mask.
Parameters
- self : the dnn module
- i_thread_index : thread index for which affinity mask will be set
- i_affinity_mask : an affinity mask represented in hex ("0x..") or binary ("0b..") form.
morapi_Dnn_getAffinityMaskNum()
morapi_Int32 morapi_Dnn_getAffinityMaskNum (const morapi_Dnn * self)
Gets the number of affinity masks.
Parameters
- self : the dnn module
morapi_Dnn_getAffinityMask()
const morapi_Char * morapi_Dnn_getAffinityMask (morapi_Dnn * self,
morapi_Int32 i_thread_index)
Gets an affinity mask in hex ("0x..) form.
Parameters
- self : the dnn module
- i_thread_index : thread index for which affinity mask will be get
morapi_Dnn_resetThreadNumAndAffinityMask()
morapi_Result morapi_Dnn_resetThreadNumAndAffinityMask (morapi_Dnn * self)
Resets the number of threads and affinity mask to the default.
Parameters
- self : the dnn module.
morapi_Dnn_setBoosted()
morapi_Result morapi_Dnn_setBoosted (morapi_Dnn * self,
morapi_Bool i_is_boosted)
Sets boosted multi-threads (MORAPI_TRUE) or not (MORAPI_FALSE).
Boosted multi-thread mode allows to 'busy wait' the other threads. The default is MORAPI_TRUE.
Parameters
- self : the dnn module
- i_is_boosted : whether boosted or not.
morapi_Dnn_isBoosted()
morapi_Bool morapi_Dnn_isBoosted (const morapi_Dnn * self)
Gets whether boosted multi-thread mode or not.
Parameters
- self : the dnn module
Returns
whether boosted or not.
morapi_Dnn_setBatchSize()
morapi_Result morapi_Dnn_setBatchSize (morapi_Dnn * self,
morapi_Int32 i_batch_size)
Sets the batch size for forward propagation.
Parameters
- self : the dnn module
- i_batch_size : batch size
morapi_Dnn_getBatchSize()
morapi_Int32 morapi_Dnn_getBatchSize (const morapi_Dnn * self)
Gets the batch size for forward propagation.
Parameters
- self : the dnn module
morapi_Dnn_setInputShape()
morapi_Result morapi_Dnn_setInputShape (morapi_Dnn * self,
morapi_Int32 i_input_index,
morapi_Int32 i_rank,
const morapi_Int32 * i_shape)
Sets the i_input_index-th input shape.
Parameters
- self : the dnn module
- i_input_index : input index
- i_rank : input rank
- i_shape : shape to be set
morapi_Dnn_setBufOptEnabled()
morapi_Result morapi_Dnn_setBufOptEnabled (morapi_Dnn * self,
const morapi_Device * i_device,
morapi_Bool i_is_enabled)
Enables or disables buffer optimization.
Buffer optimization reduces memory usage during forward propagation.
Parameters
- self : the dnn module
- i_device : device for which buffer optimization will be enabled/disabled
- i_is_enabled : boolean indicating whether it will be enabled (true) or disabled (false)
morapi_Dnn_isBufOptEnabled()
morapi_Bool morapi_Dnn_isBufOptEnabled (const morapi_Dnn * self,
const morapi_Device * i_device)
Returns whether buffer optimization is enabled or not.
Parameters
- self : the dnn module
- i_device : device to be checked
morapi_Dnn_setDumpEnabled()
morapi_Result morapi_Dnn_setDumpEnabled (morapi_Dnn * self,
morapi_Bool i_is_dump_enabled)
Enables or disables data dump of layer outputs by morapi_Dnn_dump().
Precondition
Call this before morapi_Dnn_compile(). This method automatically uncompiles if already compiled.
Parameters
- self : the dnn module
- i_is_dump_enabled : false to disable dump, true to enable it
morapi_Dnn_isDumpEnabled()
morapi_Bool morapi_Dnn_isDumpEnabled (const morapi_Dnn * self)
Returns whether data dump is enabled or not.
Parameters
- self : the dnn module
morapi_Dnn_setDumpDir()
morapi_Result morapi_Dnn_setDumpDir (morapi_Dnn * self,
const morapi_Char * i_dump_dir)
Sets the dump directory.
If NULL or empty, no file is dumped.
Parameters
- self : the dnn module
- i_dump_dir : dump directory
morapi_Dnn_setModelText()
morapi_Result morapi_Dnn_setModelText (morapi_Dnn * self,
const morapi_Char * i_text)
Sets the model text.
Parameters
- self : the dnn module
- i_text : text to be set
morapi_Dnn_getModelTextSize()
morapi_Int32 morapi_Dnn_getModelTextSize (morapi_Dnn * self)
Gets the length of JSON text data for the Dnn network structure.
Parameters
- self : the dnn module
morapi_Dnn_getModelText()
morapi_Result morapi_Dnn_getModelText (morapi_Dnn * self,
morapi_Char * o_text,
morapi_Int32 i_text_size)
Gets the JSON text data for Dnn network structure.
Parameters
- self : the dnn module
- o_text : output model text
- i_text_size : text size
morapi_Dnn_addTensor()
morapi_Tensor * morapi_Dnn_addTensor (morapi_Dnn * self,
const morapi_Char * i_name)
Adds a new tensor to the Dnn module.
Parameters
- self : the dnn module
- i_name : tensor name
morapi_Dnn_removeTensor()
morapi_Result morapi_Dnn_removeTensor (morapi_Dnn * self,
morapi_Tensor * io_tensor)
Removes the tensor from the Dnn.
Parameters
- self : the dnn module
- io_tensor : tensor to be removed
morapi_Dnn_getTensorNum()
morapi_Int32 morapi_Dnn_getTensorNum (const morapi_Dnn * self)
Gets the number of named tensors stored in the Dnn.
Parameters
- self : the dnn module
morapi_Dnn_getTensor()
morapi_Tensor * morapi_Dnn_getTensor (const morapi_Dnn * self,
morapi_Int32 i_index)
Gets the i_index-th named tensor in the Dnn.
Parameters
- self : the dnn module
- i_index : tensor index
morapi_Dnn_findTensor()
morapi_Tensor * morapi_Dnn_findTensor (const morapi_Dnn * self,
const morapi_Char * i_name)
Finds a tensor in the Dnn by name.
Parameters
- self : the dnn module
- i_name : tensor name
morapi_Dnn_getNetNum()
morapi_Int32 morapi_Dnn_getNetNum (const morapi_Dnn * self)
Gets the number of DnnNet subnetworks in the Dnn.
Parameters
- self : the dnn module
Returns
the number of DnnNet
morapi_Dnn_getNet()
morapi_DnnNet * morapi_Dnn_getNet (const morapi_Dnn * self,
morapi_Int32 i_index)
Gets the i_index-th DnnNet subnetwork in the Dnn.
Parameters
- self : the dnn module
- i_index : net index
Returns
the i_index-th DnnNet
morapi_Dnn_findMainNet()
morapi_DnnNet * morapi_Dnn_findMainNet (const morapi_Dnn * self)
Finds the main subnetwork in the Dnn.
Parameters
- self : the dnn module
Returns
the biggest subnetwork
morapi_Dnn_addNet()
morapi_DnnNet * morapi_Dnn_addNet (morapi_Dnn * self,
const morapi_Char * i_name)
Adds a net.
Parameters
- self : the dnn module
- i_name : net name
Returns
the resulted DnnNet module
morapi_Dnn_removeNet()
morapi_Result morapi_Dnn_removeNet (morapi_Dnn * self,
morapi_DnnNet * io_net)
Removes the net.
Parameters
- self : the dnn module
- io_net : net to be removed
morapi_Dnn_findNet()
morapi_DnnNet * morapi_Dnn_findNet (const morapi_Dnn * self,
const morapi_Char * i_name)
Finds the net named i_name.
Parameters
- self : the dnn module
- i_name : name of the net to be found
Returns
pointer to the DnnNet module
morapi_Dnn_parse()
morapi_Result morapi_Dnn_parse (morapi_Dnn * self)
Parses the dnn.
Parameters
- self : the dnn module
morapi_Dnn_compile()
morapi_Result morapi_Dnn_compile (morapi_Dnn * self)
Compiles the dnn to run forward propagation.
Parameters
- self : the dnn module
morapi_Dnn_cleanup()
morapi_Result morapi_Dnn_cleanup (morapi_Dnn * self)
Cleans up.
Parameters
- self : the dnn module to be cleaned up
morapi_Dnn_getState()
morapi_DnnState morapi_Dnn_getState (const morapi_Dnn * self)
Gets the state.
Parameters
- self : the dnn module
morapi_Dnn_isFailed()
morapi_Bool morapi_Dnn_isFailed (const morapi_Dnn * self)
Returns whether self is failed or not.
Parameters
- self : the dnn module
morapi_Dnn_clear()
morapi_Result morapi_Dnn_clear (morapi_Dnn * self)
Removes all tensors and nets.
Parameters
- self : the dnn module
morapi_Dnn_isEmpty()
morapi_Bool morapi_Dnn_isEmpty (const morapi_Dnn * self)
Returns whether self is empty or not.
morapi_Dnn_loadFromMemory()
morapi_Result morapi_Dnn_loadFromMemory (morapi_Dnn * self,
morapi_Uint8 * i_model_buf,
morapi_Int64 i_model_size)
load the dnn from memory.
Parameters
- self : the dnn module
- i_model_buf : input model buffer
- i_model_size : input model size
morapi_Dnn_isSecret()
morapi_Bool morapi_Dnn_isSecret (const morapi_Dnn * self)
Returns whether the dnn is secret or not.
Parameters
- self : the dnn module
morapi_Dnn_isPasswordProtected()
morapi_Bool morapi_Dnn_isPasswordProtected (const morapi_Dnn * self)
Returns whether self is password-protected or not.
morapi_Dnn_isLazy()
morapi_Bool morapi_Dnn_isLazy (const morapi_Dnn * self)
Returns whether the dnn is lazy loaded or not.
Parameters
- self : the dnn module
morapi_Dnn_setBinaryLibEmbedded()
morapi_Bool morapi_Dnn_setBinaryLibEmbedded (morapi_Dnn * self,
morapi_Bool i_is_embedded)
Sets whether the binary lib is embedded when serialization.
Parameters
- self : the dnn object.
- i_is_embedded : whether the binary lib is embedded or not.
morapi_Dnn_isBinaryLibEmbedded()
morapi_Bool morapi_Dnn_isBinaryLibEmbedded (const morapi_Dnn * self)
Returns whether the binary lib is embedded when serialization.
Parameters
- self : the dnn object.
morapi_Dnn_saveBinaryLib()
morapi_Result morapi_Dnn_saveBinaryLib (morapi_Dnn * self,
const morapi_Char * i_file_name,
morapi_Bool i_saves_only_when_changed)
Saves the binary lib to a file.
Parameters
- self : the dnn object.
- i_file_name : the file name.
- i_saves_only_when_changed : whether to save the binary library only when it is changed.
morapi_Dnn_loadBinaryLib()
morapi_Result morapi_Dnn_loadBinaryLib (morapi_Dnn * self,
const morapi_Char * i_file_name)
Loads the binary lib from a file.
morapi_Dnn_tune()
morapi_Result morapi_Dnn_tune (morapi_Dnn * self,
const morapi_DnnRecipe * i_recipe)
Tune the dnn by the tuning recipe.
Parameters
- self : the dnn module
- i_recipe : tuning recipe
morapi_Dnn_isTuned()
morapi_Bool morapi_Dnn_isTuned (const morapi_Dnn * self)
Returns whether the dnn is tuned or not.
Parameters
- self : the dnn module
morapi_Dnn_getTuneParams()
const morapi_Params * morapi_Dnn_getTuneParams (const morapi_Dnn * self)
Gets tuning parameters.
morapi_Dnn_recycle()
morapi_Result morapi_Dnn_recycle (morapi_Dnn * self)
Recycles the dnn by decompiling and unparsing.
Parameters
- self : the dnn module
morapi_Dnn_getInput()
morapi_Tensor * morapi_Dnn_getInput (const morapi_Dnn * self,
morapi_Int32 i_input_index)
Gets the i_input_index-th input tensor.
Parameters
- self : the dnn module
- i_input_index : input tensor index
morapi_Dnn_getOutput()
morapi_Tensor * morapi_Dnn_getOutput (const morapi_Dnn * self,
morapi_Int32 i_output_index)
Gets the i_output_index-th output tensor.
Parameters
- self : the dnn module
- i_output_index : output tensor index
morapi_Dnn_reset()
morapi_Result morapi_Dnn_reset (morapi_Dnn * self)
Resets the network state for cases where weights change during execution.
Parameters
- self : the dnn module
morapi_Dnn_squashNet()
morapi_Result morapi_Dnn_squashNet (morapi_Dnn * self,
morapi_DnnNet * io_net,
const morapi_Char * i_name)
Squashes the net.
Friends And Related Function Documentation
morapi_Dnn_static_installRoutineSchema()
morapi_Result morapi_Dnn_static_installRoutineSchema (morapi_Env * env,
const morapi_Char * i_schema)
Installs a routine schema.
morapi_Dnn_static_getRoutineSchemaNum()
morapi_Int32 morapi_Dnn_static_getRoutineSchemaNum (morapi_Env * env)
Gets the number of installed routine schemas.
Parameters
- env : environment where the routine are installed.
morapi_Dnn_static_getRoutineSchemas()
const morapi_Char ** morapi_Dnn_static_getRoutineSchemas (morapi_Env * env)
Gets the installed routine schemas.
Parameters
- env : environment where the routines are installed.