Intel® oneAPI Deep Neural Network Developer Guide and Reference
A newer version of this document is available. Customers should click here to go to the newest version.
Visible to Intel only — GUID: GUID-7B6988D1-C03D-42E9-9001-0AD6DB8F0BE9
Visible to Intel only — GUID: GUID-7B6988D1-C03D-42E9-9001-0AD6DB8F0BE9
enum dnnl_accumulation_mode_t
Overview
Accumulation mode. More…
#include <dnnl_common_types.h> enum dnnl_accumulation_mode_t { dnnl_accumulation_mode_strict, dnnl_accumulation_mode_relaxed, dnnl_accumulation_mode_any, dnnl_accumulation_mode_s32, dnnl_accumulation_mode_f32, dnnl_accumulation_mode_f16, };
Detailed Documentation
Accumulation mode.
Enum Values
dnnl_accumulation_mode_strict
Default behavior, f32/f64 for floating point computation, s32 for integer.
dnnl_accumulation_mode_relaxed
Same as strict but allows some partial accumulators to be rounded to src/dst datatype in memory.
dnnl_accumulation_mode_any
uses fastest implementation, could use src/dst datatype or wider datatype for accumulators
dnnl_accumulation_mode_s32
use s32 accumulators during computation
dnnl_accumulation_mode_f32
use f32 accumulators during computation
dnnl_accumulation_mode_f16
use f16 accumulators during computation