24.04
|
Go to the documentation of this file.
24 #ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG_H
25 #define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG_H
72 return _configs.at(
DT_F32);
74 return _configs.at(
DT_F16);
85 std::array<T, 3> _configs;
117 #endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG_H
@ QSYMM8_PER_CHANNEL
quantized, symmetric per channel fixed-point 8-bit number
@ QASYMM8
quantized, asymmetric fixed-point 8-bit number unsigned
virtual ~IClMatMulNativeKernelConfig()=default
Virtual destructor.
Basic interface for the matmul native kernel configuration This is the base class that chooses archit...
static constexpr size_t DT_F16
Alias for F16 index.
T get_function(DataType data_type)
Method to return the matmul native configuration function based on data type.
IClMatMulNativeKernelConfig(GPUTarget arch)
Constructor.
static constexpr size_t DT_INT8
Alias for Int8 index.
ClMatMulNativeConfigArray(T func_f32, T func_f16, T func_int8)
Constructor.
@ QASYMM8_SIGNED
quantized, asymmetric fixed-point 8-bit number signed
GPUTarget
Available GPU Targets.
Basic container for the OpenCL MatMul Native configuration functions.
static constexpr size_t DT_F32
Alias for F32 index.
Copyright (c) 2017-2024 Arm Limited.
@ F16
16-bit floating-point number
ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(IClMatMulNativeKernelConfig)
Store the tensor's metadata.
@ F32
32-bit floating-point number
virtual MatMulKernelInfo configure(const ITensorInfo *lhs, const ITensorInfo *rhs, const MatMulInfo &info)=0
This method returns the MatMulKernelInfo for the given inputs.
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
DataType
Available data types.
Class for holding information related to matrix multiplication function.