Code

Address issues of top-k op#16670

Code

Draft

asandhupatlaTT wants to merge 16 commits intomaintenstorrent/tt-metal:mainfrom asandhupatla/13235tenstorrent/tt-metal:asandhupatla/13235Copy head branch name to clipboard

+684-137

Commits

Commits on Jan 22, 2025

• add skeleton support for largest & sorted

  Show description for c8c2e3b
  asandhupatlaTT

  committed
  c8c2e3b

  Copy full SHA for c8c2e3b

• potential patch for k = 64

  Show description for 3d67320
  asandhupatlaTT

  committed
  3d67320

  Copy full SHA for 3d67320

• k=32 passes but for k=64 faiks: patch1

  Show description for bf1a99b
  asandhupatlaTT

  committed
  bf1a99b

  Copy full SHA for bf1a99b

• missing one last sort in rebuild

  Show description for 212eb7b
  asandhupatlaTT

  committed
  212eb7b

  Copy full SHA for 212eb7b

• passes for all cases except when W=8192

  Show description for 5a65a8f
  asandhupatlaTT

  committed
  5a65a8f

  Copy full SHA for 5a65a8f

• add test cases for testing

  Show description for 953d389
  asandhupatlaTT

  committed
  953d389

  Copy full SHA for 953d389

• fix bug when we rebuild same tile twice at same time

  Show description for 7a0ee2d
  asandhupatlaTT

  committed
  7a0ee2d

  Copy full SHA for 7a0ee2d

• single core top-k works for k=64

  Show description for d146cc0
  asandhupatlaTT

  committed
  d146cc0

  Copy full SHA for d146cc0

• patch 1 for multicore support

  Show description for e9288d1
  asandhupatlaTT

  committed
  e9288d1

  Copy full SHA for e9288d1

• fix bug where we get wrong writer pointer for k > 32

  Show description for 6743b03
  asandhupatlaTT

  committed
  6743b03

  Copy full SHA for 6743b03

• support all dims

  Show description for cab6201
  asandhupatlaTT

  committed
  cab6201

  Copy full SHA for cab6201

• fix test cases of top-k

  Show description for 67772b1
  asandhupatlaTT

  committed
  67772b1

  Copy full SHA for 67772b1

• cleanup

  Show description for 71d4360
  asandhupatlaTT

  committed
  71d4360

  Copy full SHA for 71d4360

• add sorted flags to test case

  Show description for 4f86077
  asandhupatlaTT

  committed
  4f86077

  Copy full SHA for 4f86077

• support K which is not power of 2

  Show description for 1f3d380
  asandhupatlaTT

  committed
  1f3d380

  Copy full SHA for 1f3d380

• support n-d tensor

  Show description for 9d4eef9
  asandhupatlaTT

  committed
  9d4eef9

  Copy full SHA for 9d4eef9