# AscendC_910B_Operators

**Repository Path**: liushuo_ustc/AscendC_910B_Operators

## Basic Information

- **Project Name**: AscendC_910B_Operators
- **Description**: [蓝区开发] 个人蓝区->黄区同步用仓库
AscendC 基于910B芯片 开发的算子
- **Primary Language**: C++
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 5
- **Created**: 2023-11-24
- **Last Updated**: 2024-07-07

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# AscendC 910 Operators (Matmul_All)
#### 适用硬件
Ascend 910(B), Ascend 310p

#### 软件架构
当前基于的CANN版本：
https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN_Test/20231215/Ascend-cann-toolkit_CANN-7.1_linux-x86_64.run

https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN_Test/20231215/Ascend-cann-toolkit_CANN-7.1_linux-aarch64.run

https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN_Test/20231215/Ascend-cann-communitysdk_CANN-7.1_linux-x86_64.run

https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN_Test/20231215/Ascend-cann-communitysdk_CANN-7.1_linux-aarch64.run

python版本最好在 3.7.5及以上

#### 算子清单
**====== MatMulAll ======**

Input	inputX	Tensor	FP16	ND	[M, K]

Input	inputWeight	Tensor	FP16	ND	[K, N]

Input	inputBias	Tensor	FP16	ND	[N]

Input	inputExtra	Tensor	FP16	ND	[M, N]

Output	outMatrix	Tensor	FP16	ND	[M, N]

Attrs   needTrans   Scalar  Int     None   1/0

Attrs   withBias   Scalar  Int     None   1/0

Attrs   operateType   Scalar  Int     None   0/1/2/3/4(None/+/-/*//)

#### 安装算子前的设置

如果在ascend910B3上安装则跳过。

根据硬件型号，设置[build_ascendc_ops.sh](.build/build_ascendc_ops.sh)中`${msopgen} gen -i ir_demo.json -f onnx \
        -c ai_core-{soc-version} -lan cpp -out ${dst}`为对应的硬件型号。目前我们在ascend910B3上出现了编译的问题。


#### 使用说明

```
cd build

bash build_ops.sh

./custom_project/build_out/custom_opp_${OS}_aarch64.run

cd ../tests/ascendc/matmul_all/

bash run_static.sh
```
> 可以根据需要配置cmake编译选项，如`cmake .. -DDYNAMIC_SHAPE=ON`表示采用动态调用方式进行编译(TODO)