Eigen::TensorSycl::internal::ScanLauncher_impl< CoeffReturnType, stp > Struct Template Reference

#include <TensorScanSycl.h>

Static Public Member Functions
template<typename Input , typename EvaluatorPointerType , typename Reducer , typename Index >
static EIGEN_STRONG_INLINE void	scan_block (Input in_ptr, EvaluatorPointerType out_ptr, Reducer &accumulator, const Index total_size, const Index scan_size, const Index panel_size, const Index non_scan_size, const Index scan_stride, const Index non_scan_stride, const bool inclusive, const Eigen::SyclDevice &dev)

Member Function Documentation

scan_block()

template<typename CoeffReturnType , scan_step stp>

template<typename Input , typename EvaluatorPointerType , typename Reducer , typename Index >

static EIGEN_STRONG_INLINE void Eigen::TensorSycl::internal::ScanLauncher_impl< CoeffReturnType, stp >::scan_block ( Input  in_ptr, EvaluatorPointerType  out_ptr, Reducer &  accumulator, const Index  total_size, const Index  scan_size, const Index  panel_size, const Index  non_scan_size, const Index  scan_stride, const Index  non_scan_stride, const bool  inclusive, const Eigen::SyclDevice &  dev  )

inlinestatic

                                                                         {
    auto scan_info =
        ScanInfo<Index>(total_size, scan_size, panel_size, non_scan_size, scan_stride, non_scan_stride, dev);
    const Index temp_pointer_size = scan_info.block_size * non_scan_size * panel_size;
    const Index scratch_size = scan_info.max_elements_per_block / (ScanParameters<Index>::ScanPerThread / 2);
    CoeffReturnType *temp_pointer =
        static_cast<CoeffReturnType *>(dev.allocate_temp(temp_pointer_size * sizeof(CoeffReturnType)));
    EvaluatorPointerType tmp_global_accessor = dev.get(temp_pointer);
 
    typedef ScanKernelFunctor<Input, CoeffReturnType, EvaluatorPointerType, Reducer, Index, stp> ScanFunctor;
    dev.template binary_kernel_launcher<CoeffReturnType, ScanFunctor>(
           in_ptr, out_ptr, tmp_global_accessor, scan_info.get_thread_range(), scratch_size,
           scan_info.get_scan_parameter(), accumulator, inclusive)
        .wait();
 
    if (scan_info.block_size > 1) {
      ScanLauncher_impl<CoeffReturnType, scan_step::second>::scan_block(
          tmp_global_accessor, tmp_global_accessor, accumulator, temp_pointer_size, scan_info.block_size, panel_size,
          non_scan_size, Index(1), scan_info.block_size, false, dev);
 
      SYCLAdjustBlockOffset<EvaluatorPointerType, CoeffReturnType, Reducer, Index>::adjust_scan_block_offset(
          tmp_global_accessor, out_ptr, accumulator, total_size, scan_size, panel_size, non_scan_size, scan_stride,
          non_scan_stride, dev);
    }
    dev.deallocate_temp(temp_pointer);
  }

References Eigen::TensorSycl::internal::SYCLAdjustBlockOffset< EvaluatorPointerType, CoeffReturnType, Reducer, Index >::adjust_scan_block_offset().

Referenced by Eigen::internal::ScanLauncher< Self, Reducer, Eigen::SyclDevice, vectorize >::operator()().

---

The documentation for this struct was generated from the following file:

• TensorScanSycl.h