mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-06-29 11:16:59 +00:00
78d657c4f7
[CK_TILE] Integrate CK Tile Dispatcher code generation into CK Tile Profiler (#7284) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit ## Motivation CK Tile is going to be delivered to hipDNN via CK Dispatcher. Currently the CK Tile Profiler using CK Builder for generating the profiled instances from the configuration files that identify the instances that old CK exposes. We need to replace this instance generation with the CK Tile Dispatcher codegen. ## Technical Details The old CK Profiler config files are converted to JSON files that the CK Tile Dispatcher can digest. The conversion script for configurations is stored to source control in case we need to update the JSON configurations later. The dispatcher generates instance libraries per conv direction (fwd, bwd data, and bwd weight) that are linked to the CK Profiler executable. I also implemented codegne for the stream-K and depthwise conv instances. The proposed solution replaces the CK Builder codegen with the CK Tile Dispatcher codegen. There are two new methods that are exposed via the dispatcher backend - `is_supported` - required to enabled the profiler workflow where we check the applicability of the kernel instance before running it. - `get_instance_string` - this mainly for verification. This provide the CK Builder instance string for verifying that the old CK Builder based profiler and the new CK Tile Dispatcher based profiler have the same instances. The rules that limit the generated instances are now collected to a single location under the dispacther. The CK Builder codegen uses these, which ensures that the two codegen pipelines are in sync. The next step (different PR) is to remove the CK Builder codegen pipeline altogether. ## Test Plan Verified that the old CK Builder based profiler and the new CK Tile Dispatcher based profiler have the same instances, that is, the Dispatcher based codgen can generate the same instances as the old CK Builder. ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
CK Tile Unified Code Generators
Single source of truth for GEMM and Grouped Convolution kernel generation.
See also: Main Dispatcher README for installation and core concepts.
Shared Infrastructure
Both GEMM and Grouped Conv generators share common code via codegen_common.py:
TileConfig- Dataclass for tile dimensionsTraitConfigBase- Base for kernel trait configurations with arch-aware validationCommonTypeMappings- Dtype-to-C++ type mappingsparallel_generate()- Parallel kernel generation with per-kernel progress logging- Arch-aware expansion helpers (
valid_wave_configs,valid_warp_configs, etc.)
Quick Start
GEMM
cd dispatcher/codegen
# Generate standard FP16 kernels
python3 unified_gemm_codegen.py \
--output-dir ../build/generated_kernels \
--datatype fp16 \
--layout rcr \
--variants standard
# Generate all variants
python3 unified_gemm_codegen.py \
--output-dir ../build/generated_kernels \
--variants standard preshuffle multi_d
Grouped Convolution
cd dispatcher/codegen
# Generate forward FP16 grouped conv kernels
python3 unified_grouped_conv_codegen.py \
--output-dir ../build/generated_kernels \
--datatype fp16 \
--variant forward \
--ndim-spatial 2
# Generate backward data kernels
python3 unified_grouped_conv_codegen.py \
--output-dir ../build/generated_kernels \
--variant backward_data \
--ndim-spatial 2
Using from Python
from ctypes_utils import CodegenRunner, KernelConfig
# Generate from specific config
config = KernelConfig(tile_m=256, tile_n=256, tile_k=64)
codegen = CodegenRunner()
result = codegen.generate_from_config(config)
# Generate variant
result = codegen.generate("preshuffle")
# Generate all
results = codegen.generate_all()
Command Line Options
| Option | Values | Description |
|---|---|---|
--output-dir |
path | Output directory |
--datatype |
fp16, bf16, fp32, int8 |
Data type |
--layout |
rcr, rrr, crr, ccr |
Matrix layouts |
--gpu-target |
gfx942, gfx90a, gfx950 |
Target GPU |
--variants |
standard, preshuffle, multi_d |
Kernel variants |
--preselected |
fp16_rcr_essential, etc. |
Predefined kernel set |
Layout Notation
R= Row-major,C= Column-major- Order: A, B, C (e.g.,
rcr= A row, B col, C row)
Variants
Standard
Basic GEMM: C = A x B
PreShuffle
Optimized weight access with LDS pre-shuffling. Best for large matrices.
Multi-D
Element-wise fusion: C = op(A x B + D0 + D1 + ...)
Supported ops: PassThrough, MultiDAdd, Relu, Gelu, Sigmoid, Tanh
Output Structure
generated_kernels/
|---- gemm_fp16_rcr_compv4_..._128x128x32_....hpp # GEMM kernels
|---- gemm_fp16_rcr_compv4_..._preshuffle.hpp
|---- gemm_fp16_rcr_compv4_..._multid_Relu_d1.hpp
|---- grouped_conv_fwd_fp16_nhwgc_..._128x128x32_....hpp # Grouped conv kernels
+---- ...
Configuration Files
arch_specs.json
GPU architecture specifications (single source of truth):
{
"architectures": {
"gfx942": {
"family": "cdna3",
"warp_size": 64,
"warp_configs": [[2, 2, 1], [4, 4, 1]],
...
}
}
}
preselected_kernels.py
Curated kernel sets for common use cases.
Adding New GPU Support
See ADDING_NEW_GPU.md for complete guide.
Quick steps:
- Edit
arch_specs.json - Run
python generate_arch_specs.py - Rebuild
Troubleshooting
| Issue | Solution |
|---|---|
| "Arguments not supported" | Check tile config validity |
| Missing element-wise op | Check elementwise_ops.hpp |
| Compilation errors | Verify C++17, include paths |
More info: See ../README.md for full documentation.