1. 30 Apr, 2017 2 commits
  2. 27 Apr, 2017 1 commit
    • Ondrej Mosnáček's avatar
      Convert CUDA kernel to warp-shuffle style · 5ce99e0d
      Ondrej Mosnáček authored
      This commit converts the CUDA kernel to a style similar to the code in
      the warp-shuffle branch (shared memory is used to simulate the warp
      shuffles). This makes it use 4 times less shared memory and some more
      registers (but not too much).
      5ce99e0d
  3. 10 Apr, 2017 1 commit
  4. 05 Apr, 2017 1 commit
  5. 04 Apr, 2017 1 commit
  6. 31 Mar, 2017 2 commits
  7. 21 Mar, 2017 1 commit
  8. 15 Mar, 2017 2 commits
    • Ondrej Mosnáček's avatar
      Pre-fill memory for tuning · fec58846
      Ondrej Mosnáček authored
      With Argon2d, when the tuning was run over uninitialized memory, the
      computation was much faster than usual. Most likely, the memory was initialized
      to zeroes by CUDA runtime, therefore causing predictable (and cache-friendly)
      memory accesses when reference positions were computed from all-zero blocks.
      
      This commit fixes this behavior by pre-initializing the memory with a NULL
      password before tuning.
      fec58846
    • Ondrej Mosnáček's avatar
      Minor fix in debug output formatting · b75356f9
      Ondrej Mosnáček authored
      b75356f9
  9. 14 Mar, 2017 1 commit
  10. 13 Mar, 2017 2 commits
  11. 11 Mar, 2017 2 commits
  12. 17 Feb, 2017 1 commit
  13. 16 Feb, 2017 3 commits
  14. 02 Feb, 2017 3 commits
  15. 01 Feb, 2017 2 commits
  16. 30 Jan, 2017 3 commits
  17. 27 Jan, 2017 1 commit