mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-01 12:11:19 +00:00

Files

linqunAMD c254f3d7b4 [CK_TILE] Refine Generic2dBlockShape to fix ck_tile example 2,10,11,14 on rdna3 and 4 (#2795 )

BlockWarps, WarpTile in Generic2dBlockShape are wave size dependent, it causes mangled name mismatch between host and device side.

Solution: Replace them with ThreadPerBlock and move BlockWarps, WarpTile calculation into Generic2dBlockShape

2025-09-10 08:29:20 +08:00

instances

Support Wave32 in CK_TILE - Part 1 (#2594 )

2025-08-18 10:08:31 -07:00

script

[Ck_tile] smoothquant (#1617 )

2024-11-01 13:51:56 +08:00

add_rmsnorm2d_rdquant_fwd.cpp

Fix latest AITER failure and add more AITER tests in CK CI. (#2782 )

2025-09-04 13:44:00 -07:00

add_rmsnorm2d_rdquant_fwd.hpp

[CK_TILE] Refine Generic2dBlockShape to fix ck_tile example 2,10,11,14 on rdna3 and 4 (#2795 )

2025-09-10 08:29:20 +08:00

CMakeLists.txt

Label CMakeLists message() as DEBUG or STATUS for clean build output (#2301 )

2025-06-10 10:46:47 -07:00

example_add_rmsnorm2d_rdquant_fwd.cpp

[CK_TILE] Refine Generic2dBlockShape to fix ck_tile example 2,10,11,14 on rdna3 and 4 (#2795 )

2025-09-10 08:29:20 +08:00

README.md

Add json dump support to output details from CK/CKTile Examples. (#2551 )

2025-09-02 23:31:29 -07:00

README.md

Add + Rmsnorm2D + rowwise dynamic quantization forward

This folder contains example for add + Rmsnorm2D + rowwise dynamic quantization forward using ck_tile tile-programming implementation. Rdquant is short for rowwise dynamic quantization here.

build

# in the root of ck_tile
mkdir build && cd build
sh ../script/cmake-ck-dev.sh  ../ <arch>  # you can replace this <arch> to gfx90a, gfx942...
make tile_add_rmsnorm2d_rdquant_fwd -j`nproc`

This will result in an executable build/bin/tile_add_rmsnorm2d_rdquant_fwd

cmdline

args:
          -m    m dimension (default:3328)
          -n    n dimension (default:4096)
     -stride    stride per row, if -1 then equal to n (default:-1)
          -e    epsilon (default:1e-5)
     -save_x    save rms(invrms) or not. set to 1 in training case (default:1)
          -v    cpu validation or not (default:1)
      -kname    print kernel name or not (default:1)
       -prec    precision (default:fp16)
      -quant    precision (default:int8)
     -warmup    cold iter (default:5)
     -repeat    hot iter (default:20)
       -json    0: No Json, 1: Dump Results in Json format (default:0)
   -jsonfile    json file name to dump results (default:add_rmsnorm2d_rdquant_fwd.json)