Deep Neural Network Library (DNNL)  1.1.3
Performance library for Deep Learning
Enumerations
Queries

Enumerations

enum  dnnl_query_t {
  dnnl_query_undef = 0, dnnl_query_engine, dnnl_query_primitive_kind, dnnl_query_num_of_inputs_s32,
  dnnl_query_num_of_outputs_s32, dnnl_query_time_estimate_f64, dnnl_query_memory_consumption_s64, dnnl_query_scratchpad_engine,
  dnnl_query_impl_info_str, dnnl_query_reorder_src_engine, dnnl_query_reorder_dst_engine, dnnl_query_prop_kind,
  dnnl_query_some_d = 64, dnnl_query_op_d, dnnl_query_convolution_d, dnnl_query_deconvolution_d,
  dnnl_query_shuffle_d, dnnl_query_eltwise_d, dnnl_query_softmax_d, dnnl_query_pooling_d,
  dnnl_query_lrn_d, dnnl_query_batch_normalization_d, dnnl_query_layer_normalization_d, dnnl_query_inner_product_d,
  dnnl_query_rnn_d, dnnl_query_gemm_d, dnnl_query_binary_d, dnnl_query_some_md = 128,
  dnnl_query_src_md, dnnl_query_diff_src_md, dnnl_query_weights_md, dnnl_query_diff_weights_md,
  dnnl_query_dst_md, dnnl_query_diff_dst_md, dnnl_query_workspace_md, dnnl_query_scratchpad_md
}
 Primitive descriptor query specification. More...
 

Detailed Description

Enumeration Type Documentation

◆ dnnl_query_t

Primitive descriptor query specification.

For generic function dnnl_primitive_desc_query(), the type of result must agree with the queried argument. The correspondence table:

Query | type of result

dnnl_query_engine | dnnl_engine_t * dnnl_query_scratchpad_engine | dnnl_engine_t * dnnl_query_primitive_kind | dnnl_primitive_kind_t * *_s32 | int * *_s64 | dnnl_dim_t * (same as int64_t *) *_f64 | double * *_str | const char ** dnnl_query_op_d | const_dnnl_op_desc_t * *_md | const dnnl_memory_desc_t ** *_${op}_d | const dnnl_${op}_desc_t ** *_pd | const_dnnl_primitive_desc_t *

Note
Rule of thumb: all opaque types and structures are returned by reference. All numbers are returned by value.
Warning
All returned references point to constant objects and are valid only during the lifetime of the queried primitive descriptor. Returned objects must not be destroyed by the user. If you need to keep the object longer than the lifetime of the queried primitive descriptor, use dnnl_primitive_desc_clone() to make a copy.
Enumerator
dnnl_query_undef 

no query

dnnl_query_engine 

execution engine

dnnl_query_primitive_kind 

primitive kind

dnnl_query_num_of_inputs_s32 

number of inputs expected

dnnl_query_num_of_outputs_s32 

number of outputs expected

dnnl_query_time_estimate_f64 

runtime estimation (seconds)

dnnl_query_memory_consumption_s64 

memory consumption – extra

dnnl_query_scratchpad_engine 

(scratch) memory, additional to all inputs and outputs memory (bytes)

scratchpad engine – engine to be used

dnnl_query_impl_info_str 

for creating scratchpad memory

implementation name

dnnl_query_reorder_src_engine 

source engine

dnnl_query_reorder_dst_engine 

destination engine

dnnl_query_prop_kind 

propagation kind

dnnl_query_some_d 

stub

dnnl_query_op_d 

op descriptor

dnnl_query_convolution_d 

convolution descriptor

dnnl_query_deconvolution_d 

deconvolution descriptor

dnnl_query_shuffle_d 

shuffle descriptor

dnnl_query_eltwise_d 

eltwise descriptor

dnnl_query_softmax_d 

softmax descriptor

dnnl_query_pooling_d 

pooling descriptor

dnnl_query_lrn_d 

lrn descriptor

dnnl_query_batch_normalization_d 

batch normalization descriptor

dnnl_query_layer_normalization_d 

layer normalization descriptor

dnnl_query_inner_product_d 

inner product descriptor

dnnl_query_rnn_d 

rnn descriptor

dnnl_query_gemm_d 

GEMM descriptor.

dnnl_query_binary_d 

binary descriptor

dnnl_query_some_md 

stub

dnnl_query_src_md 

source memory desc

dnnl_query_diff_src_md 

source gradient memory desc

dnnl_query_weights_md 

weights memory descriptor desc

dnnl_query_diff_weights_md 

weights grad. memory desc

dnnl_query_dst_md 

destination memory desc

dnnl_query_diff_dst_md 

destination grad. memory desc

dnnl_query_workspace_md 

workspace memory desc

dnnl_query_scratchpad_md 

scratchpad memory desc