mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-20 21:09:08 +00:00
945849b0f5
[CK_TILE] Add depthwise conv2d forward kernel (FP16/FP32) (#6838) ## Motivation CK currently has no kernel optimized for depthwise convolution (G=C_in=C_out, C=K=1 per group) and existing generic paths perform poorly for this workload. This PR adds a dedicated depthwise conv forward kernel in CK Tile. ## Technical Details Adds a dedicated depthwise conv2d forward op to CK Tile that performs direct convolution rather than falling back to the generic GEMM path. The kernel is templatized by filter size, stride, and data type, and compiled into ~60 instances covering common configurations (kernel 3/5/7/9, stride 1/2, FP16/FP32). Supports both CDNA (gfx942/gfx950) and RDNA (gfx1100/gfx1200) architectures. ## Test Plan - [x] Correctness and performance validated on gfx942, gfx950, and gfx1100, with ckProfiler `grouped_conv_fwd` as baseline. - [ ] MI300A (gfx942) and gfx1200 validation. ## Submission Checklist - [x ] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. AICK-1137
21 lines
1.3 KiB
C++
21 lines
1.3 KiB
C++
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
|
// SPDX-License-Identifier: MIT
|
|
#pragma once
|
|
|
|
#include "ck_tile/ops/grouped_convolution/kernel/grouped_convolution_backward_data_kernel.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/kernel/grouped_convolution_backward_weight_kernel.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/kernel/grouped_convolution_forward_kernel.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/pipeline/grouped_conv_universal_pipeline_ag_bg_cr_policy.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/pipeline/grouped_convolution_forward_depthwise_pipeline.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/utils/convolution_specialization.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/utils/grouped_convolution_utils.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/utils/split_k_utils.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/utils/transform_conv_bwd_data_to_gemm.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/utils/transform_conv_bwd_weight_to_gemm.hpp"
|
|
#include "ck_tile/ops/grouped_convolution/utils/transform_conv_fwd_to_gemm.hpp"
|
|
#include "ck_tile/ops/common/generic_2d_block_shape.hpp"
|
|
#include "ck_tile/ops/common/load_and_convert_tile.hpp"
|
|
#include "ck_tile/ops/common/streamk_common.hpp"
|
|
#include "ck_tile/ops/common/tensor_layout.hpp"
|
|
#include "ck_tile/ops/common/utils.hpp"
|