Function BlockJacobiTransKernel

ngscuda/dev_blockjacobi.cpp:69–86

    

  __global__ void BlockJacobiTransKernel (double s, FlatArray<Dev<BlockJacobiCtr>> ctrs, 
                                          BareVector<Dev<double>> x, BareVector<Dev<double>> y)
  {
    for (int i = blockIdx.x*blockDim.y+threadIdx.y; i < ctrs.Size(); i += gridDim.x*blockDim.y)
      {
        BlockJacobiCtr mv = ctrs[i];
        size_t h = mv.mat.Height();
        size_t w = mv.mat.Width();
     
        for (int r = threadIdx.x; r < w; r += blockDim.x)
          {
            double sum = 0;
            for (int c = 0; c < h; c++)
              sum += mv.mat(c,r) * x(mv.indices[c]);
            atomicAdd((double*)&y(mv.indices[r]), s*sum);
          }
      }
  }