Advertisement
Guest User

Untitled

a guest
Jun 30th, 2017
468
0
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 1.91 KB | None | 0 0
  1. commit 80174bfdd30da508576d318323af57b2cd821c5b
  2. Author: Oskar Arvidsson <oskar@irock.se>
  3. Date: Sat Oct 30 20:16:33 2010 +0200
  4.  
  5. x86 asm for high-bit-depth pixel metrics
  6. Overall speed change from these 6 asm patches: ~4.4x.
  7. But there's still tons more asm to do -- patches welcome!
  8.  
  9. Breakdown from this patch:
  10. ~13x faster SAD than C.
  11. ~11.5x faster SATD than C (only MMX done).
  12. ~18.5x faster SA8D than C.
  13. ~19.2x faster hadamard_ac than C.
  14. ~8.3x faster SSD than C.
  15. ~12.4x faster VAR than C.
  16. ~3-4.2x faster intra SAD than C.
  17. ~7.9x faster intra SATD than C.
  18.  
  19. commit 3b818190c22f7ba6d3b2f228954b1368f58f75dc
  20. Author: Jason Garrett-Glaser <darkshikari@gmail.com>
  21. Date: Sat Oct 30 19:13:05 2010 -0700
  22.  
  23. x86 asm for some high-bit-depth coefficient functions
  24. ~7.9x faster denoise than C.
  25. ~2.3x faster coeff_level_run than C.
  26. ~6.6x faster coeff_last than C.
  27. ~4.3x faster decimate_score than C.
  28.  
  29. Also improve checkasm's decimate_score test.
  30.  
  31. commit faed8d5a8ca66d0904a43b545376832bc4a038c9
  32. Author: Oskar Arvidsson <oskar@irock.se>
  33. Date: Sun Nov 14 03:33:30 2010 +0100
  34.  
  35. x86 asm for high-bit-depth motion compensation
  36. ~8x faster qpel MC than C.
  37. ~10x faster hpel than C.
  38.  
  39. commit 38861b90aa958e16743920271f5a4f509d12c6ff
  40. Author: Oskar Arvidsson <oskar@irock.se>
  41. Date: Thu Nov 4 02:13:43 2010 +0100
  42.  
  43. x86 asm for high-bit-depth quant
  44. ~3.1-4.2x faster than C.
  45.  
  46. commit 82a3dc116318d5594ac3474112cfe3472ca1b31e
  47. Author: Oskar Arvidsson <oskar@irock.se>
  48. Date: Sat Oct 30 16:55:48 2010 +0200
  49.  
  50. x86 asm for high-bit-depth DCT
  51. Only MMX and DCT done so far; iDCT still needs asm as well.
  52. ~4.4x faster than C.
  53.  
  54. commit 2584843007df862ecf99e1d357f7ef617c582184
  55. Author: Oskar Arvidsson <oskar@irock.se>
  56. Date: Sat Oct 30 11:42:52 2010 +0200
  57.  
  58. x86 asm for high-bit-depth deblocking
  59. ~3.3x faster than C.
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement