Represents memory formats (orders).
In CNN most of data is described as 4 dimensional blocks. In Intel(R) clDNN library we describe memory with 4 letters. More...

#include <tensor.hpp>

Public Types
enum	type : int32_t { yxfb = cldnn_format_yxfb, byxf = cldnn_format_byxf, bfyx = cldnn_format_bfyx, fyxb = cldnn_format_fyxb, os_iyx_osv16 = cldnn_format_os_iyx_osv16, bs_xs_xsv8_bsv8 = cldnn_format_bs_xs_xsv8_bsv8, bs_xs_xsv8_bsv16 = cldnn_format_bs_xs_xsv8_bsv16, bs_x_bsv16 = cldnn_format_bs_x_bsv16, bf8_xy16 = cldnn_format_bf8_xy16, image_2d_weights_c4_fyx_b = cldnn_format_image_2d_weights_c4_fyx_b, image_2d_weights_c1_b_fyx = cldnn_format_image_2d_weights_c1_b_fyx, winograd_2x3_s1_data, winograd_2x3_s1_weights, winograd_2x3_s1_fused_weights, format_num = cldnn_format_format_num, any = cldnn_format_any }

Public Member Functions
size_t	batch_num () const
	Returns number of batch dimensions.

size_t	feature_num () const
	Returns number of feature dimensions.

size_t	spatial_num () const
	Returns number of spatial dimensions.

const std::string &	order () const
	Returns an order of dimensions in form of string.

const std::string &	internal_order () const
	Returns an internal orders of dimensions form of string.

size_t	dimension () const
	Returns number of dimensions contained within this format.

bool	is_winograd () const
	Checks if `format` is a winograd format.

bool	is_image_2d () const
	Checks if `format` is of image 2d type.

bool	is_image () const
	Checks if `format` is of image type.

constexpr	format (type t)
	Implicit conversion from format::type.

constexpr	operator type () const
	Implicit conversion to format::type.

constexpr	format (cldnn_format_type t)
	Conversion from C API cldnn_format_type.

constexpr	operator cldnn_format_type () const
	Conversion to C API cldnn_format_type.

Static Public Member Functions
static const format_traits &	traits (type fmt)
	Get format traits for particular `format::type`.

static size_t	batch_num (type fmt)
	Returns number of batch dimensions for a `format`.

static size_t	feature_num (type fmt)
	Returns number of feature dimensions for a `format`.

static size_t	spatial_num (type fmt)
	Returns number of spatial dimensions for a `format`.

static const std::string &	order (type fmt)
	Returns an order of dimensions for a @ format.

static const std::string &	internal_order (type fmt)
	Returns an internal orders of dimensions for a `format`.

static size_t	dimension (type fmt)
	Returns number of dimensions contained within a `format`.

static bool	is_winograd (type fmt)
	Checks if `format` is a winograd format.

static bool	is_image_2d (type fmt)
	Checks if `format` is of image2d type.

static bool	is_image (type fmt)
	Checks if `format` is of image type.

Public Attributes
type	value

Detailed Description

Represents memory formats (orders).
In CNN most of data is described as 4 dimensional blocks. In Intel(R) clDNN library we describe memory with 4 letters.

b - number of blocks in batch. For weights formats: output features - conv, neurons - inner product
f - number of feature maps, features or channels. For weights formats: input features - conv, inputs, inner product
x - spatial, width
y - spatial, height /n For explanation how each format type is implemented in memory we will use naming shown bellow (b=2,f=3,y=3,x=3):

Definition at line 75 of file tensor.hpp.

Member Enumeration Documentation

◆ type

enum cldnn::format::type : int32_t

Enumerator
yxfb	batch first, feature and than spatials
byxf	used in bitmaps, input from user i.e b images of RGB format
bfyx	the most common format for activations in clDNN.
fyxb	format not used inside clDNN, but supported in reorder as extension for user provided formats.
os_iyx_osv16	format used only for convolution weights: os - output feature maps slice, i - input feature maps, yx - spatials, sv16 - 16 values of single slice.
bs_xs_xsv8_bsv8	format used only for fully connected weights: bs - batch slice, xs - x slice, bsv8 - 8 values of single slice.
bs_xs_xsv8_bsv16	format used only for fully connected weights: bs - batch slice, xs - x slice, bsv16 - 16 values of single slice.
bs_x_bsv16	format used only for fully connected weights fp16 batch=1 : bs - batch slice (responses slice), bsv16 - 16 values of single batch slice, x - flattened plane of (fyx).
bf8_xy16	format used only for convolution 1x1 input, xy aligned to 16, f aligned to 8
image_2d_weights_c4_fyx_b	image format for weights, width size is fyx/4 (4-channels filled with fyx data), height is b
image_2d_weights_c1_b_fyx	image format for weights, width size is b, height is fyx, single channel
winograd_2x3_s1_data	format used for input for winograd convolution, F(2,3) – filter 3x3 with stride 1
winograd_2x3_s1_weights	format used for weights for winograd non-fused convolution, F(2,3) – filter 3x3 with stride 1
winograd_2x3_s1_fused_weights	format used for weights for winograd fused convolution, F(2,3) – filter 3x3 with stride 1
format_num	number of format types