| 27_matmul_gelu接入msTuner[示例]
Co-authored-by: init__zhb__<zhanghaobo6@huawei.com>
# message auto-generated for no-merge-commit merge:
!467 merge pacr/mmgelu_tuner into master
27_matmul_gelu接入msTuner[示例]
Created-by: init__zhb__
Commit-by: init__zhb__
Merged-by: cann-robot
Description: ## 描述
将27号样例MatmulGelu算子接入msTuner框架。
## 关联的Issue
https://gitcode.com/cann/catlass/issues/130
## 原因
性能调优需要，外加带Epilogue的msTuner接入动作。
## 测试
自编译、运行验证通过。
```bash
# build & compile
bash scripts/build.sh --clean -DCATLASS_LIBRARY_KERNELS=27_matmul_gelu mstuner_catlass
# Run test
export LD_LIBRARY_PATH=$PWD/output/lib64/:$LD_LIBRARY_PATH
./output/bin/mstuner_catlass --m=512 --n=1024 --k=4096 --accu_dtype=fp32 --device=1 --output=result.csv
```
tuner结果：
```bash
Top 10:
case_id,task_duration(us),device_id,operation,description,m,n,k,A,B,C,element_size
1790,44.361,1,Gemm,catlass_gemm_27_matmul_gelu_fp16xRowMajor_fp16xRowMajor_fp16xRowMajor_112x256x256_112x256x32_swizzle3x0,512,1024,4096,fp16:row,fp16:row,fp32:row,4
1791,44.641,1,Gemm,catlass_gemm_27_matmul_gelu_fp16xRowMajor_fp16xRowMajor_fp16xRowMajor_112x256x256_112x256x48_swizzle3x0,512,1024,4096,fp16:row,fp16:row,fp32:row,4
1792,45.081,1,Gemm,catlass_gemm_27_matmul_gelu_fp16xRowMajor_fp16xRowMajor_fp16xRowMajor_112x256x256_112x256x64_swizzle3x0,512,1024,4096,fp16:row,fp16:row,fp32:row,4
```
## 文档更新
## 类型标签
- [ ] Bug修复
- [x] 新特性
- [ ] 性能优化
- [ ] 文档更新
- [ ] 其他，请描述：
See merge request: cann/catlass!467 | 2 个月前 |
