Binary¶
General¶
The binary primitive computes the result of a binary elementwise operation between tensors source 0 and source 1 (the variable names follow the standard Naming Conventions):
where \(op\) is one of the following operators: addition (\(+\)), subtraction (\(-\)), multiplication (\(\times\)), division (\(\div\)), greater than or equal to (\(\geq\)), greater than (\(>\)), less than or equal to (\(\leq\)), less than (\(<\)), equal to (\(=\)), not equal to (\(\neq\)), get maximum value (\(\max(\cdot)\)), get minimum value (\(\min(\cdot)\)), and conditional select operation. For the conditional select operation, the binary primitive uses a third input tensor \(src_2\) to select between the two source tensors:
The binary primitive does not have a notion of forward or backward propagations.
Execution Arguments¶
When executed, the inputs and outputs should be mapped to an execution argument index as specified by the following table.
Primitive input/output |
Execution argument index |
---|---|
\(\src_0\) |
DNNL_ARG_SRC_0 |
\(\src_1\) |
DNNL_ARG_SRC_1 |
\(\src_2\) |
DNNL_ARG_SRC_2 |
\(\dst\) |
DNNL_ARG_DST |
\(\text{binary post-op}\) |
DNNL_ARG_ATTR_MULTIPLE_POST_OP(binary_post_op_position) | DNNL_ARG_SRC_1 |
\(binary scale0\) |
DNNL_ARG_ATTR_SCALES | DNNL_ARG_SRC_0 |
\(binary scale1\) |
DNNL_ARG_ATTR_SCALES | DNNL_ARG_SRC_1 |
Implementation Details¶
General Notes¶
The binary primitive requires all source and destination tensors to have the same number of dimensions.
The binary primitive supports implicit broadcast semantics for source 0 and source 1. This means that if a dimension size is one, that single value will be broadcast (used to compute an operation with each point of the other source) for that dimension. It is recommended to use broadcast for source 1 to get better performance. Generally it should match the syntax below:
{N,1}x{C,1}x{D,1}x{H,1}x{W,1}:{N,1}x{C,1}x{D,1}x{H,1}x{W,1} -> NxCxDxHxW
. It is consistent with PyTorch broadcast semantic.The dimensions of both sources must match unless either is equal to one.
\(\src_1\) and \(\dst\) memory formats can be either specified explicitly or by dnnl::memory::format_tag::any (recommended), in which case the primitive will derive the most appropriate memory format based on the format of the source 0 tensor. The \(\dst\) tensor dimensions must match the ones of the source 0 and source 1 tensors (except for broadcast dimensions).
The binary primitive supports in-place operations, meaning that source 0 tensor may be used as the destination, in which case its data will be overwritten. In-place mode requires the \(\dst\) and source 0 data types to be the same. Different data types will unavoidably lead to correctness issues.
For the binary select operation, broadcast semantics are not supported for the third conditional input tensor. For this case, the dimensions and layout of the conditional input tensor must match that of the source 0 tensor.
Post-Ops and Attributes¶
The following attributes are supported:
Type |
Operation |
Description |
Res |
---|---|---|---|
Attribute |
Scales the corresponding input tensor by the given scale factor(s). |
Only one scale per tensor is supported. Input tensors only. |
|
Post-op |
Adds the operation result to the destination tensor instead of overwriting it. |
||
Post-op |
Applies an Eltwise operation to the result. |
||
Post-op |
Applies a Binary operation to the result |
General binary post-op restrictions |
Data Types Support¶
The source and destination tensors may have f32
, bf16
, f16
, s32
or s8/u8
data types. For the binary select operation, the conditional input tensor can only be of s8
data type. The binary primitive supports the following combinations of data types:
Source 0 / 1 |
Destination |
---|---|
f32, bf16, f16, s32, u8, s8 |
f32, bf16, f16, s32, u8, s8 |
Warning
There might be hardware and/or implementation specific restrictions. Check Implementation Limitations section below.
Data Representation¶
Sources, Destination¶
The binary primitive works with arbitrary data tensors. There is no special meaning associated with any of tensors dimensions.
Implementation Limitations¶
Refer to Data Types for limitations related to data types support.
GPU
Only tensors of 6 or fewer dimensions are supported.
s32 data type is not supported.
Performance Tips¶
Whenever possible, avoid specifying different memory formats for source tensors.
Examples¶
This C++ API example demonstrates how to create and execute a Binary primitive.
Key optimizations included in this example:
In-place primitive execution;
Primitive attributes with fused post-ops.
Bnorm u8 by Binary Post-Ops Example
Bnorm u8 via binary postops example.