* 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.
*/
#include <ATen/Operators.h>
#include <torch/all.h>
#include <torch/library.h>
#include "acl/acl.h"
#include "torch_npu/csrc/core/npu/NPUStream.h"
#include "torch_npu/csrc/core/npu/DeviceUtils.h"
#include "torch_npu/csrc/framework/OpCommand.h"
#include "tiling/platform/platform_ascendc.h"
#include "op_kernel/CustVec.h"
#include "kernel_operator.h"
namespace npu_ops_transformer_ext {
namespace Mambav2Rmsnormgated {
class CubeHandler{
public:
__aicore__ inline CubeHandler(){}
__aicore__ inline void Init(GM_ADDR xmtx, GM_ADDR zmtx, GM_ADDR wmtx, GM_ADDR outmtx, GM_ADDR workspace, int B, int S, int D, int G, float E){
int offset = 0;
}
__aicore__ inline void Compute(){
}
private:
TPipe pipe;
};
class VecHandler{
public:
__aicore__ inline VecHandler(){}
__aicore__ inline void Init(GM_ADDR xmtx, GM_ADDR zmtx, GM_ADDR wmtx, GM_ADDR outmtx, GM_ADDR workspace, int B, int S, int D, int G, float E){
int offset = 0;
tilingShapeCustVec(B, S, D, G, E, vec0shape);
vec0.Init(xmtx, zmtx, wmtx, outmtx, vec0shape);
}
__aicore__ inline void Compute(){
vec0.Compute();
}
private:
TPipe pipe;
CustVec vec0;
CustVecShapeInfo vec0shape;
};
__global__ __aicore__ void kernel_cust_rmsnormgated(GM_ADDR xmtx, GM_ADDR zmtx, GM_ADDR wmtx, GM_ADDR outmtx, GM_ADDR workspace, int B, int S, int D, int G, float E){
if ASCEND_IS_AIC{
CubeHandler cube;
cube.Init(xmtx, zmtx, wmtx, outmtx, workspace, B, S, D, G, E);
cube.Compute();
mad((__cc__ float*)0, (__ca__ float*)0, (__cb__ float*)0, 32, 32, 32, 0x0, false, false, true);
}
if ASCEND_IS_AIV{
VecHandler vec;
vec.Init(xmtx, zmtx, wmtx, outmtx, workspace, B, S, D, G, E);
vec.Compute();
copy_ubuf_to_ubuf((__ubuf__ float*)0, (__ubuf__ float*)0, 0, 1, 1, 0, 0);
}
}
at::Tensor mambav2_rmsnormgated(at::Tensor &x, at::Tensor &z, at::Tensor &w, int64_t G, double E){
auto ascendcPlatform = platform_ascendc::PlatformAscendCManager::GetInstance();
uint32_t blockDims = 20;
int devidx = x.device().index();
c10_npu::set_device(devidx);
int B = x.size(0);
int S = x.size(1);
int D = x.size(2);
at::Tensor x_fp32 = x.to(at::kFloat);
at::Tensor z_fp32 = z.to(at::kFloat);
at::Tensor w_fp32 = w.to(at::kFloat);
c10_npu::NPUStream stream = c10_npu::getCurrentNPUStream(devidx);
void* aclstream = stream.stream();
at::Tensor output = torch::empty(x.sizes(), x.options());
auto userWorkspaceSize = 1024;
size_t systemWorkspaceSize = static_cast<size_t>(ascendcPlatform->GetLibApiWorkSpaceSize());
int64_t workspaceSize = userWorkspaceSize + systemWorkspaceSize;
auto workspaceTensor = at::empty({workspaceSize}, at::TensorOptions().dtype(at::kByte).device(x.options().device()));
auto acl_call = [=]() -> int {
kernel_cust_rmsnormgated<<<blockDims, nullptr, aclstream>>>(
(GM_ADDR)(x_fp32.storage().data()),
(GM_ADDR)(z_fp32.storage().data()),
(GM_ADDR)(w_fp32.storage().data()),
(GM_ADDR)(output.storage().data()),
(GM_ADDR)(workspaceTensor.storage().data()),
B, S, D, G, E
);
return 0;
};
at_npu::native::OpCommand::RunOpApiV2("Mambav2Rmsnormgated", acl_call);
return output;
}
torch::Tensor mambav2_rmsnormgated_meta(at::Tensor &x, at::Tensor &z, at::Tensor &w, int64_t G, double E)
{
TORCH_CHECK(x.defined(), "Input tensor must be defined");
return x;
}
TORCH_LIBRARY_IMPL(npu_ops_transformer_ext, PrivateUse1, m)
{
m.impl("mambav2_rmsnormgated", mambav2_rmsnormgated);
}
TORCH_LIBRARY_IMPL(npu_ops_transformer_ext, Meta, m)
{
m.impl("mambav2_rmsnormgated", TORCH_FN(mambav2_rmsnormgated_meta));
}
}
}