Method configure

src/cpu/kernels/CpuMulKernel.cpp:1560–1727

} // namespace

void CpuMulKernel::configure(ITensorInfo   *src1,
                             ITensorInfo   *src2,
                             ITensorInfo   *dst,
                             float          scale,
                             ConvertPolicy  overflow_policy,
                             RoundingPolicy rounding_policy)
{
    ARM_COMPUTE_UNUSED(rounding_policy);
    ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);

    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src1, src2, dst, scale, overflow_policy, rounding_policy));

    const TensorShape &out_shape = TensorShape::broadcast_shape(src1->tensor_shape(), src2->tensor_shape());

    // Auto initialize dst if not initialized
    set_shape_if_empty(*dst, out_shape);

    _scale          = scale;
    _scale_exponent = 0;
    _func_quantized = nullptr;
    _func_int       = nullptr;
    _func_float     = nullptr;

    bool is_scale_255 = false;
    // Check and validate scaling factor
    if (std::abs(scale - scale255_constant) < 0.00001f)
    {
        is_scale_255 = true;
    }
    else
    {
        int exponent = 0;

        std::frexp(scale, &exponent);

        // Store the positive exponent. We know that we compute 1/2^n
        // Additionally we need to subtract 1 to compensate that frexp used a mantissa of 0.5
        _scale_exponent = std::abs(exponent - 1);
    }

    const DataType dt_input1 = src1->data_type();
    const DataType dt_input2 = src2->data_type();
    const DataType dt_output = dst->data_type();
    const bool     is_sat    = (overflow_policy == ConvertPolicy::SATURATE);

    switch (dt_input1)
    {
        case DataType::QASYMM8:
            if (dt_input2 == DataType::QASYMM8 && dt_output == DataType::QASYMM8)
            {
                if (mul_q8_neon_fixedpoint_possible(src1, src2, dst, scale))
                {
                    _func_quantized = &mul_q8_neon_fixedpoint<uint8_t>;
                }
                else
                {
                    _func_quantized = &mul_saturate_quantized_8<uint8_t>;
                }