24.04
|
Go to the documentation of this file.
35 namespace experimental
37 namespace dynamic_fusion
43 _export_input_to_cl_image = cl_image;
49 return _export_input_to_cl_image;
54 _export_weights_to_cl_image = cl_image;
60 return _export_weights_to_cl_image;
71 return _fast_relaxed_math;
82 return _is_fma_available;
172 if (
conv_info.depth_multiplier > 1 && settings.
n0() > 1)
180 "Weights cannot be exported to cl_image!");
190 "Biases size and number of dst feature maps should match");
211 _component_writer{std::make_unique<GpuCkwDepthwiseConv2d>(
id, tensors, attributes, settings)}
220 return _component_writer.get();
ClComponentDepthwiseConv2dSettings & n0(unsigned int n0)
Set N0: number of columns processed by each thread.
SimpleTensor< float > src
bool export_to_cl_image(const ITensorInfo *tensor)
bool export_weights_to_cl_image() const
Get export_weights_to_cl_image flag.
unsigned int m0() const
Set M0: number of rows processed by each thread.
ClComponentDepthwiseConv2dSettings & export_input_to_cl_image(bool cl_image)
Set export_input_to_cl_image flag.
DataLayout
[DataLayout enum definition]
ClComponentDepthwiseConv2dSettings & m0(unsigned int m0)
Set M0: number of rows processed by each thread.
TensorShape compute_depthwise_convolution_shape(const ITensorInfo &input, const ITensorInfo &weights, const ConvolutionInfo &info)
Calculate the depthwise convolution output shape of a tensor.
@ NHWC
Num samples, height, width, channels.
bool export_input_to_cl_image() const
Get export_input_to_cl_image flag.
ClComponentDepthwiseConv2dSettings & export_weights_to_cl_image(bool cl_image)
Set export_weights_to_cl_image flag.
This is a generic class that packs the arguments of an operator.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
constexpr auto data_layout
Activation Layer Information class.
const IGpuCkwComponentDriver * ckw_component_driver() const override
Get writer for the component.
DepthwiseConv2dAttributes & dimension_rounding_type(const DimensionRoundingType &dimension_rounding_type)
Set Dimension rounding type.
ClComponentDepthwiseConv2d(ComponentId id, const Properties &properties, const ArgumentPack< ITensorInfo > &tensors, const Attributes &attributes, const Settings &settings)
Constructor.
Component specific settings.
Properties common to all kernel component types.
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Properties properties() const
Get properties.
bool fast_relaxed_math() const
Get fast_relaxed_math flag.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(...)
An interface used by GpuCkwDriver to write source code for a kernel component.
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
~ClComponentDepthwiseConv2d() override
Destructor.
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(t,...)
int32_t ComponentId
Uniquely identifies a kernel component within a workload.
size_t get_data_layout_dimension_index(const DataLayout &data_layout, const DataLayoutDimension &data_layout_dimension)
Get the index of the given dimension.
ClComponentDepthwiseConv2dSettings & fast_relaxed_math(bool fast_relaxed_math)
Set fast_relaxed_math flag.
DepthwiseConv2dAttributes & depth_multiplier(const uint32_t &depth_multiplier)
Set depth multiplier.
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Copyright (c) 2017-2024 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
@ F16
16-bit floating-point number
An abstract interface of a component.
DepthwiseConv2dAttributes & pad(const Padding2D &pad)
Set padding.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
static Status validate(const Properties &properties, const ArgumentPack< ITensorInfo > &tensors, const Attributes &attributes, const Settings &settings)
Validate the component.
Attributes are backend-agnostic parameters (in addition to the input/output tensors) of an operator.
@ F32
32-bit floating-point number
ArgumentPack< ITensorInfo > tensors() const
Get tensor arguments.
unsigned int n0() const
Get N0: number of columns processed by each thread.
DepthwiseConv2dAttributes & dilation(const Size2D &dilation)
Set dilation.
bool is_fma_available() const
Get is_fma_available flag.
ComponentId id() const
Get component id.
DepthwiseConv2dAttributes & stride(const Size2D &stride)
Set stride.