Forum Discussion

DongWang-BJTU's avatar
DongWang-BJTU
Icon for Occasional Contributor rankOccasional Contributor
5 years ago
Solved

Question on the new features of OpenCLSDK v20.1

Hi, I have recently noticed that in SDK v20.1 it is that the following new feature is available:

"Enabled stall enable clusters pragma to reduce local memory usage by removing FIFO instantiations and latency between functions."

however, I could not find corresponding instructions in the manual.

Has anyone tried this new feature yet ? What pragma should I use ?

Thx

  • Hi,

    It seems like there is a documentation problem as there is no information for this feature. I had reported to developer to check on this problem.

    You could able to apply this feature by using __attribute__((stall_enable)) to a kernel, then the compiler should only create stall-enable clusters in that kernel.

    Thanks

5 Replies

  • MEIYAN_L_Intel's avatar
    MEIYAN_L_Intel
    Icon for Frequent Contributor rankFrequent Contributor

    Hi,

    I had checked the document on the "stall_enable" fucntion and I am not able to find the information for this function.

    I am checking the information internally "Enabled stall enable clusters pragma to reduce local memory usage by removing FIFO instantiations and latency between functions".

    For your information, I am able to find information with HLS reference manual guide in page 148: https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/hb/hls/mnl-hls-reference.pdf

    by using the syntax "hls_use_stall_enable_clusters"

    Thanks

    • DongWang-BJTU's avatar
      DongWang-BJTU
      Icon for Occasional Contributor rankOccasional Contributor

      Hi, thanks for the quick reply.

      How do I enable this feature in OpenCL kernel code rather than HLS ? which pragma should I use ?

  • MEIYAN_L_Intel's avatar
    MEIYAN_L_Intel
    Icon for Frequent Contributor rankFrequent Contributor

    Hi,

    It seems like there is a documentation problem as there is no information for this feature. I had reported to developer to check on this problem.

    You could able to apply this feature by using __attribute__((stall_enable)) to a kernel, then the compiler should only create stall-enable clusters in that kernel.

    Thanks

  • MEIYAN_L_Intel's avatar
    MEIYAN_L_Intel
    Icon for Frequent Contributor rankFrequent Contributor

    Hi,

    I am glad that the information provided can give help to you.

    Thanks