* This program is free software, you can redistribute it and/or modify.
* Copyright (c) 2026 Huawei Technologies Co., Ltd.
* This file is a part of the CANN Open Software.
* Licensed under 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.
*/
#ifndef OPTEST_SPARSE_MATMUL_H
#define OPTEST_SPARSE_MATMUL_H
#include <torch/torch.h>
#include <tiling/platform/platform_ascendc.h>
#include "catlass_kernel_jit.h"
#include "common/run_npu_func.h"
#include "torch_utils.h"
#include "type_utils.hpp"
namespace CatlassKernelWrapper {
using SparseKernelFn = void (*)(const uint32_t, aclrtStream, const CatlassKernel::TParams&, const CatlassKernel::MatmulParams&);
template <SparseKernelFn KernelFunc>
struct SparseMatmulLike {
using OutputType = at::Tensor;
static void GetKernelInfo(
const at::Tensor& mat1, const at::Tensor& mat2, const at::Tensor& idx,
const c10::ScalarType& outDType, bool transA, bool transB,
bool formatA, bool formatB,
CatlassKernel::TParams& tParams,
CatlassKernel::MatmulParams& params)
{
tParams.element["A"] = TorchDtypeToAclDtype(mat1.scalar_type());
tParams.element["B"] = TorchDtypeToAclDtype(mat2.scalar_type());
tParams.element["C"] = TorchDtypeToAclDtype(outDType);
tParams.transpose["A"] = transA;
tParams.transpose["B"] = transB;
tParams.transpose["C"] = false;
tParams.useNz["A"] = formatA;
tParams.useNz["B"] = formatB;
tParams.useNz["C"] = false;
params.inputAddr.resize(3);
params.inputAddr[0] = static_cast<uint8_t*>(const_cast<void*>(mat1.storage().data()));
params.inputAddr[1] = static_cast<uint8_t*>(const_cast<void*>(mat2.storage().data()));
params.inputAddr[2] = static_cast<uint8_t*>(const_cast<void*>(idx.storage().data()));
int64_t m, k1, logicalK, n;
if (transA) {
m = mat1.size(1); k1 = mat1.size(0);
} else {
m = mat1.size(0); k1 = mat1.size(1);
}
n = mat2.size(1);
logicalK = mat2.size(0) * 2;
TORCH_CHECK(k1 == logicalK, "mat1 and mat2 shapes cannot be multiplied (",
m, "x", k1, " and ", logicalK, "x", n, ")");
params.m = static_cast<uint32_t>(m);
params.k = static_cast<uint32_t>(k1);
params.n = static_cast<uint32_t>(n);
}
static OutputType AllocOutput(
const CatlassKernel::TParams& tParams, CatlassKernel::MatmulParams& params)
{
OutputType output = GetOutputTensor(
{params.m, params.n}, AclDtypeToTorchDtype(tParams.elem("C")));
params.outputAddr.resize(1);
params.outputAddr[0] = static_cast<uint8_t*>(const_cast<void*>(output.storage().data()));
return output;
}
static OutputType Run(
const at::Tensor& mat1, const at::Tensor& mat2, const at::Tensor& idx,
const c10::ScalarType& outDType, bool transA, bool transB,
bool formatA, bool formatB)
{
CatlassKernel::TParams tParams;
CatlassKernel::MatmulParams params;
GetKernelInfo(mat1, mat2, idx, outDType, transA, transB, formatA, formatB, tParams, params);
OutputType output = AllocOutput(tParams, params);
aclrtStream stream = c10_npu::getCurrentNPUStream().stream(false);
uint32_t aicCoreNum = platform_ascendc::PlatformAscendCManager::GetInstance()->GetCoreNumAic();
RUN_NPU_FUNC(KernelFunc, aicCoreNum, stream, tParams, params);
return output;
}
};
}
#endif