1. 20 May, 2016 11 commits
  2. 18 May, 2016 3 commits
  3. 12 May, 2016 14 commits
  4. 09 May, 2016 1 commit
  5. 30 Apr, 2016 2 commits
    • Quentin Khan's avatar
      Adaptive kernel adaptors: fix P2L when using vectorisation · 008c4f7c
      Quentin Khan authored
      The P2L computes particles contributions using vectorized units. Each
      computation result is then reduced to the local expansion.
      
      When the particle count is not a multiple of the vector size,
      computation is done with extra zero-initialised particles. This
      behaviour is problematic with 1/r matrix kernel family when the
      interpolation point is the origin (0,0,0). The extra particles
      contribution is then Not-A-Number and the reduction propagates this to
      the full local expansion.
      
      To mitigate this issue, the last (nb_particles % vector_size) particles
      are computed one by one.
      008c4f7c
    • Quentin Khan's avatar
  6. 29 Apr, 2016 9 commits