- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
https://docs.habana.ai/en/latest/TPC/TPC_User_Guide/TPC_Coherency.html
As TPC_User_Guid stated, to ensure the coherency between each kernel, it seems that we use ASO instruction if each TPC kernel accesses the same global memory.
For instance, if we need to read and update the values in matrix C (in the same coordinates), we must block them to protect coherency, or each TPC core might read past values that haven't been updated yet.
Are there some examples of the ASO instructions?
There are no references that I could find. Currently, I have only one way to access the same global memory, which needs to consist of coherency.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
R&D pointed me at this sample: Habana_Custom_Kernel/kernels/include/pscan_combined.h at dev/zhzhang/pscan_gaudi2 · HabanaAI/Habana_Custom_Kernel
Is this sufficient?
링크가 복사됨
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Currently there are no examples of using the ASO instruction outside of the description given in the base documentation or in the TPC tutorial, given here: Use TPC Kernels on Intel® Gaudi® Technology. I will see if I can contact R&D and get a code snippet pertaining to this.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
R&D pointed me at this sample: Habana_Custom_Kernel/kernels/include/pscan_combined.h at dev/zhzhang/pscan_gaudi2 · HabanaAI/Habana_Custom_Kernel
Is this sufficient?
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Thanks, it could be a solution for me. Thanks for the reply. I really appreciate that it solved my problem.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Thanks, it could be a solution for me.
Thanks for the reply and I really appreciate that it solved my problem
