ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-04 13:41:24 +00:00
Code Issues Packages Projects Releases Wiki Activity

Ck tile/moe sorting (#1624)

Browse Source 
* add moe_sorting & check ok

* fix comments & typo

* Run remod.py under include/ck_tile & example/ck_tile directories

* format codes

* fix output ci check bug

* fix moe sorting readme and error commit file

* use magiv div to accelerate compute

* add an loop unroll for moe lds ops

* add extblocksnel to set zeros for moebufs

* [Ck_tile] moe set zero run ok, add size check and fix ref check

* [Ck_tile]fix moe_sorting fuse set_zero remod

* [Ck_tile] change name style, fix zero buffer size err, change folder

* [Ck_tile] moe_sorting: fix name style

* [Ck_tile] moe_sorting, remove useless params in traits

* [Ck_tile] change outputtile cnt * unit_size; change output buf alloc

---------

Co-authored-by: dummycoderfe <noplydummmycoder@163.com>
Co-authored-by: Po Yen, Chen <PoYen.Chen@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>
This commit is contained in:
dummycoderfe
2024-11-09 17:57:27 +08:00
committed by GitHub
parent af9546d9f4
commit bec6fbc65f
14 changed files with 770 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
include/ck_tile/ops/fused_moe/pipeline/moe_sorting_pipeline.hpp Normal file
View File

@@ -0,0 +1,39 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/ops/fused_moe/pipeline/moe_sorting_policy.hpp"
#include <string>
#include <type_traits>
#ifndef TOPK_SOFTMAX_USE_RAW_TILE_WINDOW
#define TOPK_SOFTMAX_USE_RAW_TILE_WINDOW 0
#endif
namespace ck_tile {
// template <typename Problem_, typename Policy_ = MoeSortingPolicy>
// struct MoeSortingPipeline
// {
// // TODO: this kernel only support warp per row
// using Problem = remove_cvref_t<Problem_>;
// using Policy = remove_cvref_t<Policy_>;
// using WeightType = typename Problem::WeightType;
// template <typename TopkIdWindow, typename WeightWindow>
// CK_TILE_DEVICE auto operator()(const TopkIdWindow& topk_id_window,
// const WeightWindow& weight_window,
// index_t* p_sorted_token_ids,
// WeightType* p_sorted_weights,
// index_t* p_sorted_expert_ids,
// index_t* p_total_tokens_post_pad,
// const index_t num_experts,
// const index_t unit_size,
// const size_t numel,
// const index_t topk)
// {
// }
// };
} // namespace ck_tile

15
include/ck_tile/ops/fused_moe/pipeline/moe_sorting_policy.hpp Normal file
View File

@@ -0,0 +1,15 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include "ck_tile/ops/softmax.hpp"
#include "ck_tile/ops/topk.hpp"
namespace ck_tile {
struct MoeSortingPolicy
{
};
} // namespace ck_tile

23
include/ck_tile/ops/fused_moe/pipeline/moe_sorting_problem.hpp Normal file
View File

@@ -0,0 +1,23 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
#include <string>
#include <type_traits>
namespace ck_tile {
template <typename IndexType_, typename WeightType_, index_t InternalLoadUnroll_>
struct MoeSortingProblem
{
// TODO: this kernel only support warp per row
using WeightType = remove_cvref_t<WeightType_>;
using IndexType = remove_cvref_t<IndexType_>;
static constexpr index_t WarpSize = get_warp_size();
static constexpr index_t WarpsPerBlock = 1;
static constexpr index_t InternalLoadUnroll = InternalLoadUnroll_;
};
} // namespace ck_tile