mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-03 05:01:25 +00:00
05292b3604
* Let fmha_fwd_v3() compatible with fmha_fwd() * Decouple get_fwd_blobs() and FmhaFwdKernel * Decouple compatibility checks from get_fwd_blobs() * Extract product feature checks out from get_fwd_blobs() * Remove duplicated code in factories and redundant checks * Remove FmhaFwdKernel<>::GetName() * Let FmhaFwdApiPool support pipelines with different mask_impl * Add tile setting for fmha fwd v3 pipeline * Add fwd v3 instances to tile_example_fmha_fwd manually * Remove unused function import * Undo irrelevant changes * Remove fwd v3 instances from tile_example_fmha_fwd * Finish fmha fwd v3 kernel instance codegen * Fix formatting * Remove unused F_idx attribute * Add is_generic_attention_mask<> traits * Add constraints to the fmha fwd v3 pipeline * Unify traits & problem used for fmha fwd v3 * Unify kernel launch code for fmha fwd v2 & v3 * Unify kernel template selection logic * Use same kernel codegen template for both v2 & v3 * Rename api() property as render() method * Allow specifying filter for fmha fwd api pool * Allow specifying function name when rendering api pool items * Separate fmha fwd v3 kernel dispatching logic from v2 * Remove lambda assignment * Add simple v2/v3 dispatch logic * Stop generating empty if-clauses Skip iterating over dictionaries that have no traits, and avoid assigning i_* to them. * Use "".join() to concatenate fmha fwd api string content * Add more feature checks for fmha fwd v3 pipeline * Check features before dispatch to fmha_fwd_v3() * Add more feature checks for fmha_fwd_v3() * Add missing filter call * Use Tuple to reserve the dtype orders * Fix wrong pipeline matching logic * Add fmha fwd v3 group mode instances * Add functor_transform<> * Add type constraints to make_tile_window() * Remove fmha fwd v3 example * Fix wrong product(aiter mha_fwd()) config * Fix wrong fmha fwd v2/v3 selection logic * Fix formatting * Add comment to warning v3 kernel users * Fix wrong codegen logics * Remove unnecessary param * Fix format --------- Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
CK Tile Example Suite
This directory contains a comprehensive suite of examples demonstrating the CK Tile programming model for high-performance GPU kernels. Each example illustrates a key deep learning or HPC operation, implemented using tile-based parallelism, modular pipelines, and data movement policy.
What is CK Tile?
CK Tile is a composable GPU programming API that expresses kernels as a composition of "tiles"—rectangular blocks of computation and data movement. The pipeline & policy orchestrates data movement (global <-> LDS <-> registers), computation, and synchronization, enabling high efficiency and flexibility.
Example Index
| Example | Operation | Description |
|---|---|---|
| 01_fmha | Fused Multi-Head Attention | Tile-based FMHA with masking, quantization, and epilogue fusion |
| 02_layernorm2d | LayerNorm2D | Blockwise layer normalization with fusion and quantization |
| 03_gemm | GEMM | Matrix multiplication with tilewise parallelism |
| 04_img2col | im2col | Image-to-column transformation for GEMM-based convolution |
| 05_reduce | Reduction | Tilewise sum, max, mean reductions |
| 06_permute | Permute | Generic tensor permutation (up to rank-8) |
| 09_topk_softmax | TopK-Softmax | Rowwise softmax and top-k selection for MoE gating |
| 10_rmsnorm2d | RMSNorm2D | Root mean square normalization for LLMs |
| 11_add_rmsnorm2d_rdquant | Add + RMSNorm2D + RDQuant | Fused add, RMSNorm, and rowwise dynamic quantization |
| 12_smoothquant | SmoothQuant | Per-channel scaling and quantization for int8 inference |
| 13_moe_sorting | MoE Sorting | Token-to-expert rearrangement for MoE dispatch |
| 14_moe_smoothquant | MoE-SmoothQuant | Expert-dependent quantization fused with top-k selection |
| 15_fused_moe | Fused MoE | End-to-end fused MoE block: sorting, group-GEMM, activation, weighting |
| 16_batched_gemm | Batched GEMM | Parallel computation of multiple GEMMs |
| 17_grouped_gemm | Grouped GEMM | Multiple independent GEMMs with different shapes |
| 18_flatmm | FLATMM | Flattened matrix multiplication for packed layouts |
| 19_gemm_multi_d | Multi-D GEMM | GEMM with multiple side inputs (bias, residual, etc.) |
| 35_batched_transpose | Batched Transpose | NCHW <-> NHWC and other layout conversions |
| 36_copy | Copy | Minimal example for tile-based memory movement |
| 37_transpose | Block Transpose | High-performance tiled transpose for large tensors |
Technical Highlights
- Tile Distribution: See
include/ck_tile/tile_program/tile_distribution/for mapping tiles to thread blocks. - Block Tile Pipelines: See
include/ck_tile/tile_program/block_tile_pipeline/for memory/computation pipelines. - Policies and Utilities: Many examples use custom policies for tile/block size and memory access.
How to Build & Run
mkdir build && cd build
sh ../script/cmake-ck-dev.sh ../ <arch>
make -j
Each example produces its own executable in build/bin/.
Learning and Extending
- Start Simple: Try 03_gemm or 36_copy to learn tile basics.
- Explore Fusion: See 11_add_rmsnorm2d_rdquant, 15_fused_moe, or 14_moe_smoothquant for advanced fusion.
- Experiment: Modify tile sizes, layouts, or pipelines to explore performance and flexibility.