mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-21 05:19:20 +00:00

Files

Vidyasagar Ananthan 86591de476 [rocm-libraries] ROCm/rocm-libraries#5260 (commit a1834d2)

[CK] [CK_Tile] Add FMHA scaffolding to CK kernel dispatcher
 (#5260)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Motivation

The CK Tile dispatcher currently supports GEMM and Grouped Convolution
but has no support for Fused Multi-Head Attention (FMHA). The
example/ck_tile/01_fmha folder contains a comprehensive FMHA
implementation with forward, backward, split-KV, paged-KV, append-KV,
and batch-prefill kernels across multiple GPU architectures — but there
is no unified dispatch layer for it. This PR ports the FMHA stack into
the dispatcher, following the same architectural patterns established by
GEMM and Grouped Convolution, enabling runtime kernel selection, JIT
compilation from Python, and a declarative C++ example flow. Autotuning
heuristics to follow.

## Technical Details

This PR adds FMHA scaffolding to the CK dispatcher framework, mirroring
GEMM's layered architecture. Seven new C++ runtime headers provide type
definitions (coexisting with upstream headers via __has_include,
requiring zero modifications to example/ck_tile/01_fmha/), a problem
builder with 18+ setters, Signature + Algorithm kernel key matching, a
virtual kernel instance, a DECL_FMHA_KERNEL_SET macro with wildcard
support and named tile/wave/warp setters, arch-aware registry with JSON
export, and a dispatcher with seqtune-aware selection, configurable
timing, and multi-stage execution plans for split-KV (two-stage) and
backward (three-stage). The codegen pipeline is driven by a
fmha_arch_specs.json capturing per-arch tile tables and pipeline
constraints for five architectures (gfx90a/942/950/1100/1201), migrated
from hardcoded logic in 01_fmha/codegen/, with supporting modules for
C++ symbol mappings, validation rules, and named receipt profiles
(ck_default, flash, pytorch, aiter, fp32, fp8). Python integration
(fmha_utils.py) mirrors the C++ layer with JIT compilation, parallel
multi-kernel builds, HIP memory management via ctypes, tolerance-based
validation, and a NumPy CPU reference with GQA support. Twenty-seven C++
and thirty-two Python examples cover the full feature surface — forward,
split-KV, masks, bias, dropout, GQA, backward, append-KV, batch prefill,
fp8, logits soft cap, sink tokens, and parameter sweeps — all
JIT-compiled on the fly.

## Test Plan

Seven test files cover the runtime types, codegen, and end-to-end
correctness. C++ unit tests validate the problem builder, dispatcher
planning (single-stage for forward/paged-KV/append-KV; multi-stage for
split-KV and backward), registry operations, and the kernel-set
declaration macro. Python unit tests verify codegen emission, profile
filtering, and 15 validation rules for masks, hdim constraints, and
pipeline requirements. GPU execution validation in 01_basic_fmha
--validate reports zero errors across 65,536 elements with max absolute
error of 7.29e-05. A gold-standard parity suite (test_fmha_parity.py)
runs 14 configurations through both the upstream tile_example_fmha_fwd
and the dispatcher, comparing exit codes to confirm behavioral parity —
all 14 match.

## Test Result

The C++ smoke test builds and passes all 9 compiled examples, and a
Python JIT sweep (29_sweep_seqlen.py) passes 7/7 configurations reaching
up to 375 TFLOPS at seqlen 2048.

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

2026-05-17 07:30:33 +00:00

ctypes

[rocm-libraries] ROCm/rocm-libraries#5260 (commit a1834d2)

2026-05-17 07:30:33 +00:00

README.md

[rocm-libraries] ROCm/rocm-libraries#5260 (commit a1834d2)

2026-05-17 07:30:33 +00:00

README.md

CK Tile Dispatcher - Language Bindings

This directory contains language bindings for the CK Tile Dispatcher.

Structure

bindings/
|---- ctypes/              # Python ctypes bindings (C API)
|   |---- gemm_ctypes_lib.cpp      # GEMM dispatcher C API
|   |---- conv_ctypes_lib.cpp      # Grouped conv dispatcher C API (fwd + bwd_data)
|   |---- conv_bwdw_ctypes_lib.cpp # Grouped conv backward weight C API (separate library)
|   |---- fmha_ctypes_lib.cpp      # FMHA dispatcher C API (fwd + bwd)
|   |---- gpu_helper.cpp           # CLI helper for Python
|   +---- CMakeLists.txt
+---- README.md

ctypes Bindings

The ctypes bindings provide a C API that Python can load via ctypes.CDLL().

Building

cd build
cmake .. -DCMAKE_PREFIX_PATH=/opt/rocm
make dispatcher_gemm_lib dispatcher_conv_lib gpu_helper

Usage from Python

import ctypes

# Load the library
lib = ctypes.CDLL("path/to/libdispatcher_gemm_lib.so")

# Initialize
lib.dispatcher_init()

# Check if problem is supported
is_supported = lib.dispatcher_is_supported(M, N, K)

# Run GEMM
time_ms = ctypes.c_float()
result = lib.dispatcher_run_gemm(
    A_ptr, B_ptr, C_ptr,
    M, N, K,
    ctypes.byref(time_ms)
)

# Cleanup
lib.dispatcher_cleanup()

GEMM API

Function	Description
`dispatcher_init()`	Initialize the dispatcher
`dispatcher_is_supported(M, N, K)`	Check if problem size is supported
`dispatcher_select_kernel(M, N, K, name_buf, buf_size)`	Get kernel name for problem
`dispatcher_run_gemm(A, B, C, M, N, K, time_ms)`	Execute GEMM
`dispatcher_get_kernel_count()`	Get number of registered kernels
`dispatcher_export_registry_json()`	Export registry as JSON
`dispatcher_cleanup()`	Release resources

Grouped Convolution API

Function	Description
`conv_dispatcher_init()`	Initialize the dispatcher
`conv_dispatcher_is_supported(prob)`	Check if problem is supported
`conv_dispatcher_select_kernel(prob, name_buf, buf_size)`	Get kernel name
`conv_dispatcher_run(input, weight, output, prob, stream)`	Execute convolution
`conv_dispatcher_get_kernel_count()`	Get number of registered kernels
`conv_dispatcher_cleanup()`	Release resources

GPU Helper

The gpu_helper executable provides a CLI interface for Python:

./gpu_helper 1024 1024 1024 --validate

Output is JSON for easy parsing:

{
  "problem": {"M": 1024, "N": 1024, "K": 1024},
  "kernel": "gemm_fp16_rcr_...",
  "execution": {
    "time_ms": 0.5,
    "tflops": 4.2
  },
  "validation": {
    "accuracy": 100.0
  },
  "status": "success"
}

Examples

See the examples that use these bindings:

GEMM: dispatcher/examples/gemm/python/

Grouped Convolution

Grouped convolution C++ headers and Python utilities are in:

C++ Headers: dispatcher/include/ck_tile/dispatcher/grouped_conv_*.hpp
Python Utils: dispatcher/python/grouped_conv_utils.py
Build Script: dispatcher/scripts/compile_grouped_conv_examples.py