mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-06-30 03:37:38 +00:00
33b2015939
Enable 16x16 MFMA for decode by making the softmax cross-warp reduction conditional on the warp gemm M dimension: use permlane32_swap for 32x32 MFMA (2 lanes per row), fall back to block_tile_reduce_sync for 16x16 MFMA (4 lanes per row). New tiny decode traits: 1 warp, sequence<1,1,1>, warp_gemm 16x16x32, kBlockM=16, kBlockQ=2 for GQA-8. This matches Triton's BLOCK_M=16 / BLOCK_Q=2 decode configuration exactly. Also adds 4-tier dispatch: tiny (avg_q<=2) -> small (avg_q<=8) -> medium (avg_q<=128) -> large (prefill). Benchmark results (d64 GQA-8 via aiter, 363 shapes): Before: CK faster 135 (37.2%), Triton faster 228 (62.8%) After: CK faster 247 (68.0%), Triton faster 116 (32.0%) Key shapes: 1-seq decode: 0.021ms (CK 0.75x, wins 25%) 64-seq decode: 0.025ms vs Triton 0.029ms (CK wins 14%) 512-seq decode: 0.018ms vs Triton 0.021ms (CK wins) Weighted end-to-end: CK/Triton = 0.999x (tied) Verified correct on 10 shapes: bf16+fp16, d64 GQA-8 + d128 MHA, batch 1-64, all 4 dispatch tiers. Made-with: Cursor
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.