EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Kernel] Add moe_gating_top_k operator support for Ascend NPU (#5579)

Browse Source 
### What this PR does / why we need it?

1.replace moe_gating_top_k from torch_npu with custom op
2.enable the  renorm function of moe_gating_top_k in softmax scenerio

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
No need test

- vLLM version: v0.13.0
- vLLM main:
7157596103

---------

Signed-off-by: ZCG12345 <2097562023@qq.com>
This commit is contained in:
ZCG12345
2026-01-07 21:42:31 +08:00
committed by GitHub
parent 1165b2c863
commit 3be8e33fe9
32 changed files with 4667 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

61
csrc/moe_gating_top_k/op_host/math_util.h Normal file
View File

@@ -0,0 +1,61 @@
/**
* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
/*!
* \file math_util.h
* \brief
*/
#ifndef TILING_MATMUL_MATH_UTIL_H
#define TILING_MATMUL_MATH_UTIL_H
#include <array>
#include <cstdint>
#include <vector>
#include <utility>
namespace matmul_tiling {
class MathUtil {
public:
static bool IsEqual(float leftValue, float rightValue);
template<typename T>
static auto CeilDivision(T num1, T num2) -> T
{
if (num2 == 0) {
return 0;
}
return static_cast<T>((static_cast<int64_t>(num1) + static_cast<int64_t>(num2) - 1) /
static_cast<int64_t>(num2));
}
template<typename T>
static auto Align(T num1, T num2) -> T
{
return CeilDivision(num1, num2) * num2;
}
static int32_t AlignDown(int32_t num1, int32_t num2);
static bool CheckMulOverflow(int32_t a, int32_t b, int32_t &c);
static int32_t MapShape(int32_t shape, bool roundUpFlag = true);
static void AddFactor(std::vector<int32_t> &dimsFactors, int32_t dim);
static void GetFactorCnt(const int32_t shape, int32_t &factorCnt, const int32_t factorStart,
const int32_t factorEnd);
static void GetFactorLayerCnt(const int32_t shape, int32_t &factorCnt, const int32_t factorStart,
const int32_t factorEnd);
static bool CheckFactorNumSatisfy(const int32_t dim);
static int32_t FindBestSingleCore(const int32_t oriShape, const int32_t mappedShape, const int32_t coreNum,
bool isKDim);
static void GetFactors(std::vector<int32_t> &factorList, int32_t srcNum, int32_t minFactor, int32_t maxFactor);
static void GetFactors(std::vector<int32_t> &factorList, int32_t srcNum, int32_t maxFactor);
static void GetBlockFactors(std::vector<int32_t> &factorList, const int32_t oriShape, const int32_t mpShape,
const int32_t coreNum, const int32_t maxNum);
static int32_t GetNonFactorMap(std::vector<int32_t> &factorList, int32_t srcNum, int32_t maxFactor);
static std::vector<std::pair<int, int>> GetFactorPairs(int32_t num);
static std::pair<int32_t, int32_t> DivideIntoMainAndTail(int32_t num, int32_t divisor);
};
} // namespace matmul_tiling
#endif // _MATH_UTIL_H_