0b3c297ee2
[CK Tile] Add WAVELET pipeline for forward grouped convolution (#8009) ## Motivation CK Tile forward grouped convolution trails classic CK on 3x3 convolutions whose output-channel count is not divisible by 8, where the narrow output store limits the compute CShuffle epilogue. This ports the WAVELET pipeline (added for backward-weight in #7937) to the forward kernel to close that gap. ## Technical Details - Kernel (`grouped_convolution_forward_kernel.hpp`): WAVELET load/math-wave wiring, mirroring the backward-weight implementation; the non-WAVELET path is unchanged. - Generator: implement `parse_native_fwd_instance`, the forward native-instance parser. - Registered WAVELET instances: profiler bf16 3 / fp16 5, tests 1 each. WAVELET requires input channels divisible by 8 (it does not apply to depthwise). The bf16/fp16 instance asymmetry is intentional and measured: the VecC=8 tiles never beat the compute pool in bf16 but win about 20% of divisible-by-8 3x3 shapes in fp16, so VecC=8 is registered for fp16 only. ## Test Plan - Correctness (CPU reference) for every registered profiler instance, across VecC variants. - Per-shape best-instance performance sweep over the 34 RetinaNet shapes (bf16) and a 200-shape cross-model sweep (bf16 and fp16), compared against classic CK. ## Test Result - Correctness: PASS for all instances. - RetinaNet (bf16, vs classic CK): faster on 28 of 34 shapes, geomean +19.5%; the not-divisible-by-8 shapes up to 3.7x. One 1x1 stride-2 shape stays ~20% behind classic CK, unrelated to WAVELET. - Cross-model (200 shapes): WAVELET wins 3x3 not-divisible-by-8 in both dtypes (up to 61% over the next-best compute instance); for divisible-by-8 3x3 it wins about 20% of shapes in fp16 (3-11%) and none in bf16. ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Composable Kernel Tile
concept
ck_tile provides a programming model with templated abstractions to enable users to implement performance-critical kernels for machine learning workloads. introduces following basic concepts to help users building your own operator
- tensor coordinate transformation, this is the core concept of layout/index transform abstraction in both compiler time and run time.
- tile-based programming model, including tile-level api and the concept of distributed tensor.
ck_tile is independently from the old ck, located under /include/ck_tile. You don't need to include anything from old CK, ck_tile has similiar (indeed almost the same) implementations for users to build operators. We will have a transition period to pull everything from old ck into ck_tile, stay tuned.
component
ck_tile is splitted into several componenets including core, host, ops/gemm, ops/fmha... each component you only need to include a single header (e.g #include "ck_tile/core.hpp", #include "ck_tile/ops/fmha.hpp") then you are able to use the function/structure inside (different from old ck)
[core]
ck_tile/core contains all the basic data structure and function to build the kernel, you can only include this header and build your own operators that utilizing all the basic building blocks introduced in ck.
core/container
- array, store runtime variables with fixed length (tensor index, register buffer, etc...)
- tuple, same as std::tuple, hold different type of data, and one of the solution to achieve multiple buffer.
- sequence, compile time integer sequence used to build various internal structures, or to describe tile size
- other convenient structure build on top of above 3
core/numeric
- gpu data type like
fp16_t,bf16_t,fp8_t... and the conversion between each other - constexpr integer similiar to std::integral_constant to be used as compile time integer.
- math functions and numeric utilities
core/algorithm
- coordinate transformation system, used to build tensor transform and compile time indexing. This is the core idea introduced in old
ckto describe how a tensor is build by several basic transform primitives likemerge/unmerge/embedetc... and how we indexing into a ND tensor that finally mapped to 1D memory offset.
core/tensor
- tensor descriptor, to describe how a ND tensor
- distributed tensor, describe the storage of this tensor, and the distribution of how a collection of threads collaborately work for this tensor.
- tile level API, including
load_tile,store_tile,shuffle_tile,slice_tile, etc...
[host]
ck_tile/host contains all the host side utilities to launch a kernel, create the device buffer, and some reference implementations. This can be used to create examples (like that under ck_tile example folder) and simple executable to invoke this kernel, so if you only need ck_tile to build your own device library then it's OK to not include this. Based on this, it is recommended to include the specific header you needed under this folder to avoid including unwanted headers (e.g, only include ck_tile/host/kernel_launch.hpp), unless you are writing a host executable.
[ops/gemm, ops/fmha, ops/reduce...]
our implementation of different device operators.
- warp, warp tile level operator
- block, block tile level operator
- pipeline, pipeline that can achieve a customized tile level mainloop (or epilogue). By switching different pipeline to the kernel template you can have different kind of pipeline optimizations.
- kernel, template interface for users to instantiate a particular kernel
[ops/epilogue]
epilogue part of our kernel. We may extend this epilogue part to let users to build their own cutomized epilogues.
[ref]
reference implementation of cpu or gpu. This folder is supposed to include a specific header on demand.
examples
currently we put all ck_tile related example under /example/ck_tile folder. Please check each example's subfolder.