mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-19 22:39:03 +00:00
ea10a78203
* refactor reduce kernel - Rename Reduce kernel as per convention - Move kept_dim and reduce_dims from runtime to compile-time parameters - Update Reduce2dProblem template to include KeptDim, ReduceDims, and Rank - Remove IsSupportedArgument validation function as it's unnecessary. Not using the GuaranteedLastDimensionVectorStride while making tensor view or descriptor which removes the bounds enforced earlier. We still calculate and use vector size. - Update reduce example to demonstrate NCHW->NHW reduction with non-contiguous support - Update tests Kernel now handles both contiguous and non-contiguous memory layout. * fix compile errors
Reduction with CK Tile
This example demonstrates parallel reduction (sum, max, etc.) using the CK Tile programming model, a core operation for normalization, statistics, and aggregation in deep learning.
Algorithm and Math
Given a tensor X and a reduction axis, compute:
-
Sum:
Y = \sum_i X_i -
Max:
Y = \max_i X_i -
Mean:
Y = \frac{1}{N} \sum_i X_i -
Tilewise Reduction: Each thread block reduces a tile (block) of the input, using shared memory and register accumulation for efficiency.
Tile Programming Model
- Tiles: Each thread block processes a tile (block) of the input tensor.
- Pipeline: Modular, can be extended for fused reductions or post-processing.
Build & Run
mkdir build && cd build
sh ../script/cmake-ck-dev.sh ../ <arch>
make tile_example_reduce -j
./bin/tile_example_reduce -?
Source Structure
- Kernel:
reduce.hpp(tile-programming kernel template) - Executable:
reduce.cpp(argument parsing, kernel launch) - Build:
CMakeLists.txt
Related CK Tile Examples
- 03_gemm: GEMM with tiles
- 04_img2col: im2col transformation
- 06_permute: Permutation with tiles
For distribution, see include/ck_tile/tile_program/tile_distribution/.