Batch Normalization

API Reference


The batch normalization primitive performs a forward or backward batch normalization operation on tensors with number of dimensions equal to 2 or more.


The batch normalization operation is defined by the following formulas. We show formulas only for 2D spatial data which are straightforward to generalize to cases of higher and lower dimensions. Variable names follow the standard Naming Conventions.

\[\dst(n, c, h, w) = \gamma(c) \cdot \frac{\src(n, c, h, w) - \mu(c)} {\sqrt{\sigma^2(c) + \varepsilon}} + \beta(c),\]


Mean and variance are computed at runtime or provided by a user. When mean and variance are computed at runtime, the following formulas are used:

  • \(\mu(c) = \frac{1}{NHW} \sum\limits_{nhw} \src(n, c, h, w)_{}\),

  • \(\sigma^2(c) = \frac{1}{NHW} \sum\limits_{nhw} {}_{} (\src(n, c, h, w) - \mu(c))^2\).

The \(\gamma(c)\) and \(\beta(c)\) tensors are considered learnable.

In training mode, the primitive also optionally supports fusion with ReLU activation with zero negative slope applied to the result (see dnnl_fuse_norm_relu flag).


  • The batch normalization primitive computes population mean and variance and not the sample or unbiased versions that are typically used to compute running mean and variance.

  • Using the mean and variance computed by the batch normalization primitive, running mean and variance \(\hat\mu\) and \(\hat\sigma^2\) can be computed as

    \[\begin{split}\hat\mu := \alpha \cdot \hat\mu + (1 - \alpha) \cdot \mu, \\ \hat\sigma^2 := \alpha \cdot \hat\sigma^2 + (1 - \alpha) \cdot \sigma^2.\end{split}\]

Difference Between Forward Training and Forward Inference

  • If mean and variance are computed at runtime (i.e., dnnl_use_global_stats is not set), they become outputs for the propagation kind dnnl_forward_training (because they would be required during the backward propagation) and are not exposed for the propagation kind dnnl_forward_inference.

  • If batch normalization is created with ReLU fusion (i.e., dnnl_fuse_norm_relu is set), for the propagation kind dnnl_forward_training the primitive would produce a workspace memory as one extra output. This memory is required to compute the backward propagation. When the primitive is executed with propagation kind dnnl_forward_inference, the workspace is not produced. Behavior would be the same as creating a batch normalization primitive with ReLU as a post-op (see section below).


The backward propagation computes \(\diffsrc(n, c, h, w)\), \(\diffgamma(c)^*\), and \(\diffbeta(c)^*\) based on \(\diffdst(n, c, h, w)\), \(\src(n, c, h, w)\), \(\mu(c)\), \(\sigma^2(c)\), \(\gamma(c) ^*\), and \(\beta(c) ^*\).

The tensors marked with an asterisk are used only when the primitive is configured to use \(\gamma(c)\) and \(\beta(c)\) (i.e., dnnl_use_scaleshift, dnnl_use_scale or dnnl_use_shift are set).

Execution Arguments

Depending on the flags and propagation kind, the batch normalization primitive requires different inputs and outputs. For clarity, a summary is shown below.


Inputs : \(\src\) Outputs : \(\dst\)

Inputs : \(\src\) Outputs : \(\dst\) , \(\mu\) , \(\sigma^2\)

Inputs : \(\diffdst\) , \(\src\) , \(\mu\) , \(\sigma^2\) Outputs : \(\diffsrc\)

Same as for dnnl_backward


Inputs : \(\src\) , \(\mu\) , \(\sigma^2\) Outputs : \(\dst\)

Inputs : \(\src\) , \(\mu\) , \(\sigma^2\) Outputs : \(\dst\)

Inputs : \(\diffdst\) , \(\src\) , \(\mu\) , \(\sigma^2\) Outputs : \(\diffsrc\)

Same as for dnnl_backward


Inputs : \(\src\) , \(\gamma\) , \(\beta\) Outputs : \(\dst\)

Inputs : \(\src\) , \(\gamma\) , \(\beta\) Outputs : \(\dst\) , \(\mu\) , \(\sigma^2\)

Inputs : \(\diffdst\) , \(\src\) , \(\mu\) , \(\sigma^2\) , \(\gamma\) , \(\beta\) Outputs : \(\diffsrc\) , \(\diffgamma\) , \(\diffbeta\)

Not supported


Inputs : \(\src\) , \(\gamma\) Outputs : \(\dst\)

Inputs : \(\src\) , \(\gamma\) Outputs : \(\dst\) , \(\mu\) , \(\sigma^2\)

Inputs : \(\diffdst\) , \(\src\) , \(\mu\) , \(\sigma^2\) , \(\gamma\) Outputs : \(\diffsrc\) , \(\diffgamma\)

Not supported


Inputs : \(\src\) , \(\beta\) Outputs : \(\dst\)

Inputs : \(\src\) , \(\beta\) Outputs : \(\dst\) , \(\mu\) , \(\sigma^2\)

Inputs : \(\diffdst\) , \(\src\) , \(\mu\) , \(\sigma^2\) , \(\beta\) Outputs : \(\diffsrc\) , \(\diffbeta\)

Not supported

dnnl_use_global_stats | dnnl_use_scaleshift

Inputs : \(\src\) , \(\mu\) , \(\sigma^2\) , \(\gamma\) , \(\beta\) Outputs : \(\dst\)

Inputs : \(\src\) , \(\mu\) , \(\sigma^2\) , \(\gamma\) , \(\beta\) Outputs : \(\dst\)

Inputs : \(\diffdst\) , \(\src\) , \(\mu\) , \(\sigma^2\) , \(\gamma\) , \(\beta\) Outputs : \(\diffsrc\) , \(\diffgamma\) , \(\diffbeta\)

Not supported

flags | dnnl_fuse_norm_relu

Inputs : same as with flags Outputs : same as with flags

Inputs : same as with flags Outputs : same as with flags , Workspace

Inputs : same as with flags , Workspace Outputs : same as with flags

Same as for dnnl_backward if flags do not contain dnnl_use_scaleshift ; not supported otherwise

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



\(\gamma, \beta\)






mean ( \(\mu\) )


variance ( \(\sigma^2\) )










\(\diffgamma, \diffbeta\)






Implementation Details

General Notes

  1. The different flavors of the primitive are partially controlled by the flags parameter that is passed to the operation descriptor initialization function (e.g., dnnl::batch_normalization_forward::desc::desc()). Multiple flags can be set using the bitwise OR operator (|). Flag dnnl_use_scaleshift can not be mixed with dnnl_use_scale or dnnl_use_shift.

  2. For forward propagation, the mean and variance might be either computed at runtime (in which case they are outputs of the primitive) or provided by a user (in which case they are inputs). In the latter case, a user must set the dnnl_use_global_stats flag. For the backward propagation, the mean and variance are always input parameters.

  3. The memory format and data type for src and dst are assumed to be the same, and in the API they are typically referred to as data (e.g., see data_desc in dnnl::batch_normalization_forward::desc::desc()). The same is true for diff_src and diff_dst. The corresponding memory descriptors are referred to as diff_data_desc.

  4. Both forward and backward propagation support in-place operations, meaning that \(\src\) can be used as input and output for forward propagation, and \(\diffdst\) can be used as input and output for backward propagation. In case of an in-place operation, the original data will be overwritten. Note, however, that backward propagation requires original \(\src\), hence the corresponding forward propagation should not be performed in-place.

  5. As mentioned above, the batch normalization primitive can be fused with ReLU activation even in the training mode. In this case, on the forward propagation the primitive has one additional output, workspace, that should be passed during the backward propagation.

Data Type Support

The operation supports the following combinations of data types:


Source / Destination

Mean / Variance / ScaleShift

forward / backward

f32, bf16









There might be hardware- or implementation-specific restrictions. Check the Implementation Limitations section below.

Data Representation

Mean and Variance

The mean (\(\mu\)) and variance (\(\sigma^2\)) are separate 1D tensors of size \(C\).

The format of the corresponding memory object must be dnnl_x (dnnl_a).

Scale and Shift

If dnnl_use_scaleshift is used, the scale (\(\gamma\)) and shift (\(\beta\)) are combined in a single 2D tensor of shape \(2 \times C\).

If dnnl_use_scale or dnnl_use_shift are used, the scale (\(\gamma\)) and shift (\(\beta\)) are separate 1D tensors of shape \(C\).

The format of the corresponding memory object must be dnnl_nc (dnnl_ab).

Source, Destination, and Their Gradients

Like other CNN primitives, the batch normalization primitive expects data to be \(N \times C \times SP_n \times \cdots \times SP_0\) tensor.

The batch normalization primitive is optimized for the following memory formats:


Logical tensor

Implementations optimized for memory formats



dnnl_nc ( dnnl_ab )



dnnl_ncw ( dnnl_abc ), dnnl_nwc ( dnnl_acb ), optimized^



dnnl_nchw ( dnnl_abcd ), dnnl_nhwc ( dnnl_acdb ), optimized^



dnnl_ncdhw ( dnnl_abcde ), dnnl_ndhwc ( dnnl_acdeb ), optimized^

Here optimized^ means the format that comes out of any preceding compute-intensive primitive.

Post-Ops and Attributes

Post-ops and attributes enable you to modify the behavior of the batch normalization primitive by chaining certain operations after the batch normalization operation. The following post-ops are supported by batch normalization primitives:








Applies an Eltwise operation to the result (currently only dnnl_eltwise_relu algorithm is supported)


As mentioned in Primitive Attributes, the post-ops should be used for inference only. For instance, using ReLU as a post-op would not produce the additional output workspace that is required to compute backward propagation correctly. Hence, in case of training one should use the dnnl_fuse_norm_relu directly.

Implementation Limitations

  1. Refer to Data Types for limitations related to data types support.

  2. For the data types that have forward propagation support only, mean and variance must be provided by a user (i.e., dnnl_use_global_stats is set).

  3. GPU

    • ReLU eltwise post-op doesn’t support non-zero \(\alpha\) parameter.

Performance Tips

  1. For backward propagation, use the same memory format for src, diff_dst, and diff_src (the format of the diff_dst and diff_src are always the same because of the API). Different formats are functionally supported but lead to highly suboptimal performance.

  2. Use in-place operations whenever possible (see caveats in General Notes).

  3. GPU implementations support an experimental algorithm with single pass statistics calculations. Please review experimental features for more details.


Batch Normalization Primitive Example

This C++ API example demonstrates how to create and execute a Batch Normalization primitive in forward training propagation mode.

Key optimizations included in this example:

  • In-place primitive execution;

  • Source memory format for an optimized primitive implementation;

  • Fused post-ops via operation descriptor flags;