Machine learning capabilities. More...

#include <ml_types.h>

Collaboration diagram for odp_ml_capability_t:

Data Fields
uint32_t	max_models
	Maximum number of models. More...

uint32_t	max_models_loaded
	Maximum number of models that can be loaded simultaneously.

uint64_t	max_model_size
	Maximum model size in bytes.

uint32_t	max_compl_id
	Maximum completion identifier value.

uint32_t	max_inputs
	Maximum number of model inputs.

uint32_t	max_outputs
	Maximum number of model outputs.

uint32_t	max_segs_per_input
	Maximum number of data segments per model input. More...

uint32_t	max_segs_per_output
	Maximum number of data segments per model output. More...

uint32_t	min_input_align
	Minimum input data alignment in bytes. More...

uint32_t	min_output_align
	Minimum output data alignment in bytes. More...

odp_bool_t	packed_input_data
	Input data packing. More...

odp_bool_t	packed_output_data
	Output data packing. More...

struct {
odp_ml_compl_mode_t compl_mode_mask
	Supported completion modes for model load / unload operations. More...

odp_bool_t compl_queue_plain
	Support of model load / unload completion into plain queues. More...

odp_bool_t compl_queue_sched
	Support of model load / unload completion into scheduled queues. More...

}	load
	Model load / unload capabilities.

struct {
odp_ml_compl_mode_t compl_mode_mask
	Supported completion modes for model run operations. More...

odp_bool_t compl_queue_plain
	Support of model run completion into plain queues. More...

odp_bool_t compl_queue_sched
	Support of model run completion into scheduled queues. More...

}	run
	Model run capabilities.

odp_ml_compl_pool_capability_t	pool
	ML completion event pool capabilities.

Detailed Description

Machine learning capabilities.

Examples: odp_ml_perf.c, odp_ml_run.c, and odp_sysinfo.c.

Definition at line 188 of file api/spec/ml_types.h.

Field Documentation

◆ max_models

uint32_t odp_ml_capability_t::max_models

Maximum number of models.

Maximum number of models that can be created simultaneously. The value is zero when ML offload is not available.

Definition at line 193 of file api/spec/ml_types.h.

◆ max_segs_per_input

uint32_t odp_ml_capability_t::max_segs_per_input

Maximum number of data segments per model input.

Segmented input data is not supported when 1.

Definition at line 215 of file api/spec/ml_types.h.

◆ max_segs_per_output

uint32_t odp_ml_capability_t::max_segs_per_output

Maximum number of data segments per model output.

Segmented output data is not supported when 1.

Definition at line 222 of file api/spec/ml_types.h.

◆ min_input_align

uint32_t odp_ml_capability_t::min_input_align

Minimum input data alignment in bytes.

For each model input, the first data segment must start at this or a higher power of two memory alignment in bytes. The value is 1 when there is no alignment requirement.

Definition at line 230 of file api/spec/ml_types.h.

◆ min_output_align

uint32_t odp_ml_capability_t::min_output_align

Minimum output data alignment in bytes.

For each model output, the first data segment must start at this or a higher power of two memory alignment in bytes. The value is 1 when there is no alignment requirement.

Definition at line 238 of file api/spec/ml_types.h.

◆ packed_input_data

odp_bool_t odp_ml_capability_t::packed_input_data

Input data packing.

0: Data packing is not required. 1: Data for all model inputs must be continuous in memory. The memory block starts with data for the first input and continues through all inputs in-order and without gaps between inputs. The minimum alignment requirement (min_input_align) applies only for the first input.

Definition at line 249 of file api/spec/ml_types.h.

◆ packed_output_data

odp_bool_t odp_ml_capability_t::packed_output_data

Output data packing.

0: Data packing is not required. 1: Data buffer space for all model outputs must be continuous in memory. The memory block starts with buffer space for the first output and continues through all outputs in-order and without gaps between outputs. The minimum alignment requirement (min_output_align) applies only for the first output.

Definition at line 260 of file api/spec/ml_types.h.

◆ compl_mode_mask

odp_ml_compl_mode_t odp_ml_capability_t::compl_mode_mask

Supported completion modes for model load / unload operations.

Supported completion modes for model run operations.

Mask of supported completion modes. Each supported mode has the corresponding flag (e.g. ODP_ML_COMPL_MODE_SYNC) set in the mask.

Definition at line 270 of file api/spec/ml_types.h.

◆ compl_queue_plain

odp_bool_t odp_ml_capability_t::compl_queue_plain

Support of model load / unload completion into plain queues.

Support of model run completion into plain queues.

Specifies if plain queues are supported as destination queues for load / unload completion events (ODP_ML_COMPL_MODE_EVENT).

0: Plain queues are not supported as completion queues 1: Plain queues are supported as completion queues

Specifies if plain queues are supported as destination queues for run completion events (ODP_ML_COMPL_MODE_EVENT).

0: Plain queues are not supported as completion queues 1: Plain queues are supported as completion queues

Definition at line 281 of file api/spec/ml_types.h.

◆ compl_queue_sched

odp_bool_t odp_ml_capability_t::compl_queue_sched

Support of model load / unload completion into scheduled queues.

Support of model run completion into scheduled queues.

Specifies if scheduled queues are supported as destination queues for load / unload completion events (ODP_ML_COMPL_MODE_EVENT).

0: Scheduled queues are not supported as completion queues 1: Scheduled queues are supported as completion queues

Specifies if scheduled queues are supported as destination queues for run completion events (ODP_ML_COMPL_MODE_EVENT).